xref: /titanic_52/usr/src/uts/sun4v/io/vdc.c (revision af3025fd013117424096c94bd9e116ecc538cc6b)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * LDoms virtual disk client (vdc) device driver
31  *
32  * This driver runs on a guest logical domain and communicates with the virtual
33  * disk server (vds) driver running on the service domain which is exporting
34  * virtualized "disks" to the guest logical domain.
35  *
36  * The driver can be divided into four sections:
37  *
38  * 1) generic device driver housekeeping
39  *	_init, _fini, attach, detach, ops structures, etc.
40  *
41  * 2) communication channel setup
42  *	Setup the communications link over the LDC channel that vdc uses to
43  *	talk to the vDisk server. Initialise the descriptor ring which
44  *	allows the LDC clients to transfer data via memory mappings.
45  *
46  * 3) Support exported to upper layers (filesystems, etc)
47  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
48  *	ioctl calls. vdc will copy the data to be written to the descriptor
49  *	ring or maps the buffer to store the data read by the vDisk
50  *	server into the descriptor ring. It then sends a message to the
51  *	vDisk server requesting it to complete the operation.
52  *
53  * 4) Handling responses from vDisk server.
54  *	The vDisk server will ACK some or all of the messages vdc sends to it
55  *	(this is configured during the handshake). Upon receipt of an ACK
56  *	vdc will check the descriptor ring and signal to the upper layer
57  *	code waiting on the IO.
58  */
59 
60 #include <sys/atomic.h>
61 #include <sys/conf.h>
62 #include <sys/disp.h>
63 #include <sys/ddi.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/mach_descrip.h>
69 #include <sys/modctl.h>
70 #include <sys/mdeg.h>
71 #include <sys/note.h>
72 #include <sys/open.h>
73 #include <sys/sdt.h>
74 #include <sys/stat.h>
75 #include <sys/sunddi.h>
76 #include <sys/types.h>
77 #include <sys/promif.h>
78 #include <sys/vtoc.h>
79 #include <sys/archsystm.h>
80 #include <sys/sysmacros.h>
81 
82 #include <sys/cdio.h>
83 #include <sys/dktp/fdisk.h>
84 #include <sys/dktp/dadkio.h>
85 #include <sys/scsi/generic/sense.h>
86 #include <sys/scsi/impl/uscsi.h>	/* Needed for defn of USCSICMD ioctl */
87 
88 #include <sys/ldoms.h>
89 #include <sys/ldc.h>
90 #include <sys/vio_common.h>
91 #include <sys/vio_mailbox.h>
92 #include <sys/vdsk_common.h>
93 #include <sys/vdsk_mailbox.h>
94 #include <sys/vdc.h>
95 
96 /*
97  * function prototypes
98  */
99 
100 /* standard driver functions */
101 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
102 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
103 static int	vdc_strategy(struct buf *buf);
104 static int	vdc_print(dev_t dev, char *str);
105 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
106 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
107 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
108 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
109 			cred_t *credp, int *rvalp);
110 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
111 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
112 
113 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
114 			void *arg, void **resultp);
115 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
116 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
117 
118 /* setup */
119 static void	vdc_min(struct buf *bufp);
120 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
121 static int	vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node);
122 static int	vdc_start_ldc_connection(vdc_t *vdc);
123 static int	vdc_create_device_nodes(vdc_t *vdc);
124 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
125 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
126 static int	vdc_create_device_nodes_props(vdc_t *vdc);
127 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
128 		    mde_cookie_t *vd_nodep, mde_cookie_t *vd_portp);
129 static int	vdc_get_ldc_id(md_t *, mde_cookie_t, uint64_t *);
130 static int	vdc_do_ldc_up(vdc_t *vdc);
131 static void	vdc_terminate_ldc(vdc_t *vdc);
132 static int	vdc_init_descriptor_ring(vdc_t *vdc);
133 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
134 static int	vdc_setup_devid(vdc_t *vdc);
135 static void	vdc_store_efi(vdc_t *vdc, struct dk_gpt *efi);
136 
137 /* handshake with vds */
138 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
139 static int		vdc_ver_negotiation(vdc_t *vdcp);
140 static int		vdc_init_attr_negotiation(vdc_t *vdc);
141 static int		vdc_attr_negotiation(vdc_t *vdcp);
142 static int		vdc_init_dring_negotiate(vdc_t *vdc);
143 static int		vdc_dring_negotiation(vdc_t *vdcp);
144 static int		vdc_send_rdx(vdc_t *vdcp);
145 static int		vdc_rdx_exchange(vdc_t *vdcp);
146 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
147 
148 /* processing incoming messages from vDisk server */
149 static void	vdc_process_msg_thread(vdc_t *vdc);
150 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
151 
152 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
153 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
154 static int	vdc_process_err_msg(vdc_t *vdc, vio_msg_t msg);
155 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
156 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
157 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
158 static int 	vdc_send_request(vdc_t *vdcp, int operation,
159 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
160 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
161 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
162 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
163 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
164 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
165 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation,
166 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
167 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
168 
169 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
170 static int	vdc_drain_response(vdc_t *vdcp);
171 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
172 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
173 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
174 
175 /* dkio */
176 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode);
177 static int	vdc_create_fake_geometry(vdc_t *vdc);
178 static int	vdc_setup_disk_layout(vdc_t *vdc);
179 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
180 		    int mode, int dir);
181 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
182 		    int mode, int dir);
183 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
184 		    int mode, int dir);
185 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
186 		    int mode, int dir);
187 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
188 		    int mode, int dir);
189 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
190 		    int mode, int dir);
191 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
192 		    int mode, int dir);
193 static int	vdc_uscsicmd_convert(vdc_t *vdc, void *from, void *to,
194 		    int mode, int dir);
195 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
196 		    int mode, int dir);
197 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
198 		    int mode, int dir);
199 
200 /*
201  * Module variables
202  */
203 
204 /*
205  * Tunable variables to control how long vdc waits before timing out on
206  * various operations
207  */
208 static int	vdc_retries = 10;
209 static int	vdc_hshake_retries = 3;
210 
211 static int	vdc_timeout = 0; /* units: seconds */
212 
213 /* calculated from 'vdc_usec_timeout' during attach */
214 static uint64_t	vdc_hz_timeout;				/* units: Hz */
215 static uint64_t	vdc_usec_timeout = 30 * MICROSEC;	/* 30s units: ns */
216 
217 static uint64_t vdc_hz_min_ldc_delay;
218 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
219 static uint64_t vdc_hz_max_ldc_delay;
220 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
221 
222 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
223 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
224 
225 /* values for dumping - need to run in a tighter loop */
226 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
227 static int	vdc_dump_retries = 100;
228 
229 /* Count of the number of vdc instances attached */
230 static volatile uint32_t	vdc_instance_count = 0;
231 
232 /* Soft state pointer */
233 static void	*vdc_state;
234 
235 /*
236  * Controlling the verbosity of the error/debug messages
237  *
238  * vdc_msglevel - controls level of messages
239  * vdc_matchinst - 64-bit variable where each bit corresponds
240  *                 to the vdc instance the vdc_msglevel applies.
241  */
242 int		vdc_msglevel = 0x0;
243 uint64_t	vdc_matchinst = 0ull;
244 
245 /*
246  * Supported vDisk protocol version pairs.
247  *
248  * The first array entry is the latest and preferred version.
249  */
250 static const vio_ver_t	vdc_version[] = {{1, 0}};
251 
252 static struct cb_ops vdc_cb_ops = {
253 	vdc_open,	/* cb_open */
254 	vdc_close,	/* cb_close */
255 	vdc_strategy,	/* cb_strategy */
256 	vdc_print,	/* cb_print */
257 	vdc_dump,	/* cb_dump */
258 	vdc_read,	/* cb_read */
259 	vdc_write,	/* cb_write */
260 	vdc_ioctl,	/* cb_ioctl */
261 	nodev,		/* cb_devmap */
262 	nodev,		/* cb_mmap */
263 	nodev,		/* cb_segmap */
264 	nochpoll,	/* cb_chpoll */
265 	ddi_prop_op,	/* cb_prop_op */
266 	NULL,		/* cb_str */
267 	D_MP | D_64BIT,	/* cb_flag */
268 	CB_REV,		/* cb_rev */
269 	vdc_aread,	/* cb_aread */
270 	vdc_awrite	/* cb_awrite */
271 };
272 
273 static struct dev_ops vdc_ops = {
274 	DEVO_REV,	/* devo_rev */
275 	0,		/* devo_refcnt */
276 	vdc_getinfo,	/* devo_getinfo */
277 	nulldev,	/* devo_identify */
278 	nulldev,	/* devo_probe */
279 	vdc_attach,	/* devo_attach */
280 	vdc_detach,	/* devo_detach */
281 	nodev,		/* devo_reset */
282 	&vdc_cb_ops,	/* devo_cb_ops */
283 	NULL,		/* devo_bus_ops */
284 	nulldev		/* devo_power */
285 };
286 
287 static struct modldrv modldrv = {
288 	&mod_driverops,
289 	"virtual disk client",
290 	&vdc_ops,
291 };
292 
293 static struct modlinkage modlinkage = {
294 	MODREV_1,
295 	&modldrv,
296 	NULL
297 };
298 
299 /* -------------------------------------------------------------------------- */
300 
301 /*
302  * Device Driver housekeeping and setup
303  */
304 
305 int
306 _init(void)
307 {
308 	int	status;
309 
310 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
311 		return (status);
312 	if ((status = mod_install(&modlinkage)) != 0)
313 		ddi_soft_state_fini(&vdc_state);
314 	vdc_efi_init(vd_process_ioctl);
315 	return (status);
316 }
317 
318 int
319 _info(struct modinfo *modinfop)
320 {
321 	return (mod_info(&modlinkage, modinfop));
322 }
323 
324 int
325 _fini(void)
326 {
327 	int	status;
328 
329 	if ((status = mod_remove(&modlinkage)) != 0)
330 		return (status);
331 	vdc_efi_fini();
332 	ddi_soft_state_fini(&vdc_state);
333 	return (0);
334 }
335 
336 static int
337 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
338 {
339 	_NOTE(ARGUNUSED(dip))
340 
341 	int	instance = VDCUNIT((dev_t)arg);
342 	vdc_t	*vdc = NULL;
343 
344 	switch (cmd) {
345 	case DDI_INFO_DEVT2DEVINFO:
346 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
347 			*resultp = NULL;
348 			return (DDI_FAILURE);
349 		}
350 		*resultp = vdc->dip;
351 		return (DDI_SUCCESS);
352 	case DDI_INFO_DEVT2INSTANCE:
353 		*resultp = (void *)(uintptr_t)instance;
354 		return (DDI_SUCCESS);
355 	default:
356 		*resultp = NULL;
357 		return (DDI_FAILURE);
358 	}
359 }
360 
361 static int
362 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
363 {
364 	int	instance;
365 	int	rv;
366 	vdc_t	*vdc = NULL;
367 
368 	switch (cmd) {
369 	case DDI_DETACH:
370 		/* the real work happens below */
371 		break;
372 	case DDI_SUSPEND:
373 		/* nothing to do for this non-device */
374 		return (DDI_SUCCESS);
375 	default:
376 		return (DDI_FAILURE);
377 	}
378 
379 	ASSERT(cmd == DDI_DETACH);
380 	instance = ddi_get_instance(dip);
381 	DMSGX(1, "[%d] Entered\n", instance);
382 
383 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
384 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
385 		return (DDI_FAILURE);
386 	}
387 
388 	if (vdc->open_count) {
389 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
390 		return (DDI_FAILURE);
391 	}
392 
393 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
394 
395 	/* mark instance as detaching */
396 	vdc->lifecycle	= VDC_LC_DETACHING;
397 
398 	/*
399 	 * try and disable callbacks to prevent another handshake
400 	 */
401 	rv = ldc_set_cb_mode(vdc->ldc_handle, LDC_CB_DISABLE);
402 	DMSG(vdc, 0, "callback disabled (rv=%d)\n", rv);
403 
404 	if (vdc->initialized & VDC_THREAD) {
405 		mutex_enter(&vdc->read_lock);
406 		if ((vdc->read_state == VDC_READ_WAITING) ||
407 		    (vdc->read_state == VDC_READ_RESET)) {
408 			vdc->read_state = VDC_READ_RESET;
409 			cv_signal(&vdc->read_cv);
410 		}
411 
412 		mutex_exit(&vdc->read_lock);
413 
414 		/* wake up any thread waiting for connection to come online */
415 		mutex_enter(&vdc->lock);
416 		if (vdc->state == VDC_STATE_INIT_WAITING) {
417 			DMSG(vdc, 0,
418 			    "[%d] write reset - move to resetting state...\n",
419 			    instance);
420 			vdc->state = VDC_STATE_RESETTING;
421 			cv_signal(&vdc->initwait_cv);
422 		}
423 		mutex_exit(&vdc->lock);
424 
425 		/* now wait until state transitions to VDC_STATE_DETACH */
426 		thread_join(vdc->msg_proc_thr->t_did);
427 		ASSERT(vdc->state == VDC_STATE_DETACH);
428 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
429 		    vdc->instance);
430 	}
431 
432 	mutex_enter(&vdc->lock);
433 
434 	if (vdc->initialized & VDC_DRING)
435 		vdc_destroy_descriptor_ring(vdc);
436 
437 	if (vdc->initialized & VDC_LDC)
438 		vdc_terminate_ldc(vdc);
439 
440 	mutex_exit(&vdc->lock);
441 
442 	if (vdc->initialized & VDC_MINOR) {
443 		ddi_prop_remove_all(dip);
444 		ddi_remove_minor_node(dip, NULL);
445 	}
446 
447 	if (vdc->initialized & VDC_LOCKS) {
448 		mutex_destroy(&vdc->lock);
449 		mutex_destroy(&vdc->read_lock);
450 		cv_destroy(&vdc->initwait_cv);
451 		cv_destroy(&vdc->dring_free_cv);
452 		cv_destroy(&vdc->membind_cv);
453 		cv_destroy(&vdc->sync_pending_cv);
454 		cv_destroy(&vdc->sync_blocked_cv);
455 		cv_destroy(&vdc->read_cv);
456 		cv_destroy(&vdc->running_cv);
457 	}
458 
459 	if (vdc->minfo)
460 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
461 
462 	if (vdc->cinfo)
463 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
464 
465 	if (vdc->vtoc)
466 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
467 
468 	if (vdc->label)
469 		kmem_free(vdc->label, DK_LABEL_SIZE);
470 
471 	if (vdc->devid) {
472 		ddi_devid_unregister(dip);
473 		ddi_devid_free(vdc->devid);
474 	}
475 
476 	if (vdc->initialized & VDC_SOFT_STATE)
477 		ddi_soft_state_free(vdc_state, instance);
478 
479 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
480 
481 	return (DDI_SUCCESS);
482 }
483 
484 
485 static int
486 vdc_do_attach(dev_info_t *dip)
487 {
488 	int		instance;
489 	vdc_t		*vdc = NULL;
490 	int		status;
491 	md_t		*mdp;
492 	mde_cookie_t	vd_node, vd_port;
493 
494 	ASSERT(dip != NULL);
495 
496 	instance = ddi_get_instance(dip);
497 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
498 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
499 		    instance);
500 		return (DDI_FAILURE);
501 	}
502 
503 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
504 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
505 		return (DDI_FAILURE);
506 	}
507 
508 	/*
509 	 * We assign the value to initialized in this case to zero out the
510 	 * variable and then set bits in it to indicate what has been done
511 	 */
512 	vdc->initialized = VDC_SOFT_STATE;
513 
514 	vdc_hz_timeout = drv_usectohz(vdc_usec_timeout);
515 
516 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
517 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
518 
519 	vdc->dip	= dip;
520 	vdc->instance	= instance;
521 	vdc->open_count	= 0;
522 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
523 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
524 	vdc->state	= VDC_STATE_INIT;
525 	vdc->lifecycle	= VDC_LC_ATTACHING;
526 	vdc->ldc_state	= 0;
527 	vdc->session_id = 0;
528 	vdc->block_size = DEV_BSIZE;
529 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
530 
531 	vdc->vtoc = NULL;
532 	vdc->cinfo = NULL;
533 	vdc->minfo = NULL;
534 
535 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
536 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
537 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
538 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
539 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
540 
541 	vdc->threads_pending = 0;
542 	vdc->sync_op_pending = B_FALSE;
543 	vdc->sync_op_blocked = B_FALSE;
544 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
545 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
546 
547 	/* init blocking msg read functionality */
548 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
549 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
550 	vdc->read_state = VDC_READ_IDLE;
551 
552 	vdc->initialized |= VDC_LOCKS;
553 
554 	/* get device and port MD node for this disk instance */
555 	if (vdc_get_md_node(dip, &mdp, &vd_node, &vd_port) != 0) {
556 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
557 		    instance);
558 		return (DDI_FAILURE);
559 	}
560 
561 	/* set the connection timeout */
562 	if (vd_port == NULL || (md_get_prop_val(mdp, vd_port,
563 	    VDC_MD_TIMEOUT, &vdc->ctimeout) != 0)) {
564 		vdc->ctimeout = 0;
565 	}
566 
567 	/* initialise LDC channel which will be used to communicate with vds */
568 	status = vdc_do_ldc_init(vdc, mdp, vd_node);
569 
570 	(void) md_fini_handle(mdp);
571 
572 	if (status != 0) {
573 		cmn_err(CE_NOTE, "[%d] Couldn't initialize LDC", instance);
574 		goto return_status;
575 	}
576 
577 	/* initialize the thread responsible for managing state with server */
578 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
579 	    vdc, 0, &p0, TS_RUN, minclsyspri);
580 	if (vdc->msg_proc_thr == NULL) {
581 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
582 		    instance);
583 		return (DDI_FAILURE);
584 	}
585 
586 	vdc->initialized |= VDC_THREAD;
587 
588 	atomic_inc_32(&vdc_instance_count);
589 
590 	/*
591 	 * Once the handshake is complete, we can use the DRing to send
592 	 * requests to the vDisk server to calculate the geometry and
593 	 * VTOC of the "disk"
594 	 */
595 	status = vdc_setup_disk_layout(vdc);
596 	if (status != 0) {
597 		DMSG(vdc, 0, "[%d] Failed to discover disk layout (err%d)",
598 		    vdc->instance, status);
599 		goto return_status;
600 	}
601 
602 	/*
603 	 * Now that we have the device info we can create the
604 	 * device nodes and properties
605 	 */
606 	status = vdc_create_device_nodes(vdc);
607 	if (status) {
608 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
609 		    instance);
610 		goto return_status;
611 	}
612 	status = vdc_create_device_nodes_props(vdc);
613 	if (status) {
614 		DMSG(vdc, 0, "[%d] Failed to create device nodes"
615 		    " properties (%d)", instance, status);
616 		goto return_status;
617 	}
618 
619 	/*
620 	 * Setup devid
621 	 */
622 	if (vdc_setup_devid(vdc)) {
623 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
624 	}
625 
626 	ddi_report_dev(dip);
627 	vdc->lifecycle	= VDC_LC_ONLINE;
628 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
629 
630 return_status:
631 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
632 	return (status);
633 }
634 
635 static int
636 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
637 {
638 	int	status;
639 
640 	switch (cmd) {
641 	case DDI_ATTACH:
642 		if ((status = vdc_do_attach(dip)) != 0)
643 			(void) vdc_detach(dip, DDI_DETACH);
644 		return (status);
645 	case DDI_RESUME:
646 		/* nothing to do for this non-device */
647 		return (DDI_SUCCESS);
648 	default:
649 		return (DDI_FAILURE);
650 	}
651 }
652 
653 static int
654 vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node)
655 {
656 	int			status = 0;
657 	ldc_status_t		ldc_state;
658 	ldc_attr_t		ldc_attr;
659 	uint64_t		ldc_id = 0;
660 
661 	ASSERT(vdc != NULL);
662 
663 	vdc->initialized |= VDC_LDC;
664 
665 	if ((status = vdc_get_ldc_id(mdp, vd_node, &ldc_id)) != 0) {
666 		DMSG(vdc, 0, "[%d] Failed to get LDC channel ID property",
667 		    vdc->instance);
668 		return (EIO);
669 	}
670 
671 	DMSGX(0, "[%d] LDC id is 0x%lx\n", vdc->instance, ldc_id);
672 
673 	vdc->ldc_id = ldc_id;
674 
675 	ldc_attr.devclass = LDC_DEV_BLK;
676 	ldc_attr.instance = vdc->instance;
677 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
678 	ldc_attr.mtu = VD_LDC_MTU;
679 
680 	if ((vdc->initialized & VDC_LDC_INIT) == 0) {
681 		status = ldc_init(ldc_id, &ldc_attr, &vdc->ldc_handle);
682 		if (status != 0) {
683 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
684 			    vdc->instance, ldc_id, status);
685 			return (status);
686 		}
687 		vdc->initialized |= VDC_LDC_INIT;
688 	}
689 	status = ldc_status(vdc->ldc_handle, &ldc_state);
690 	if (status != 0) {
691 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
692 		    vdc->instance, status);
693 		return (status);
694 	}
695 	vdc->ldc_state = ldc_state;
696 
697 	if ((vdc->initialized & VDC_LDC_CB) == 0) {
698 		status = ldc_reg_callback(vdc->ldc_handle, vdc_handle_cb,
699 		    (caddr_t)vdc);
700 		if (status != 0) {
701 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
702 			    vdc->instance, status);
703 			return (status);
704 		}
705 		vdc->initialized |= VDC_LDC_CB;
706 	}
707 
708 	vdc->initialized |= VDC_LDC;
709 
710 	/*
711 	 * At this stage we have initialised LDC, we will now try and open
712 	 * the connection.
713 	 */
714 	if (vdc->ldc_state == LDC_INIT) {
715 		status = ldc_open(vdc->ldc_handle);
716 		if (status != 0) {
717 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
718 			    vdc->instance, vdc->ldc_id, status);
719 			return (status);
720 		}
721 		vdc->initialized |= VDC_LDC_OPEN;
722 	}
723 
724 	return (status);
725 }
726 
727 static int
728 vdc_start_ldc_connection(vdc_t *vdc)
729 {
730 	int		status = 0;
731 
732 	ASSERT(vdc != NULL);
733 
734 	ASSERT(MUTEX_HELD(&vdc->lock));
735 
736 	status = vdc_do_ldc_up(vdc);
737 
738 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
739 
740 	return (status);
741 }
742 
743 static int
744 vdc_stop_ldc_connection(vdc_t *vdcp)
745 {
746 	int	status;
747 
748 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
749 	    vdcp->state);
750 
751 	status = ldc_down(vdcp->ldc_handle);
752 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
753 
754 	vdcp->initialized &= ~VDC_HANDSHAKE;
755 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
756 
757 	return (status);
758 }
759 
760 static int
761 vdc_create_device_nodes_efi(vdc_t *vdc)
762 {
763 	ddi_remove_minor_node(vdc->dip, "h");
764 	ddi_remove_minor_node(vdc->dip, "h,raw");
765 
766 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
767 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
768 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
769 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
770 		    vdc->instance);
771 		return (EIO);
772 	}
773 
774 	/* if any device node is created we set this flag */
775 	vdc->initialized |= VDC_MINOR;
776 
777 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
778 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
779 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
780 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
781 		    vdc->instance);
782 		return (EIO);
783 	}
784 
785 	return (0);
786 }
787 
788 static int
789 vdc_create_device_nodes_vtoc(vdc_t *vdc)
790 {
791 	ddi_remove_minor_node(vdc->dip, "wd");
792 	ddi_remove_minor_node(vdc->dip, "wd,raw");
793 
794 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
795 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
796 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
797 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
798 		    vdc->instance);
799 		return (EIO);
800 	}
801 
802 	/* if any device node is created we set this flag */
803 	vdc->initialized |= VDC_MINOR;
804 
805 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
806 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
807 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
808 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
809 		    vdc->instance);
810 		return (EIO);
811 	}
812 
813 	return (0);
814 }
815 
816 /*
817  * Function:
818  *	vdc_create_device_nodes
819  *
820  * Description:
821  *	This function creates the block and character device nodes under
822  *	/devices along with the node properties. It is called as part of
823  *	the attach(9E) of the instance during the handshake with vds after
824  *	vds has sent the attributes to vdc.
825  *
826  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
827  *	of 2 is used in keeping with the Solaris convention that slice 2
828  *	refers to a whole disk. Slices start at 'a'
829  *
830  * Parameters:
831  *	vdc 		- soft state pointer
832  *
833  * Return Values
834  *	0		- Success
835  *	EIO		- Failed to create node
836  *	EINVAL		- Unknown type of disk exported
837  */
838 static int
839 vdc_create_device_nodes(vdc_t *vdc)
840 {
841 	char		name[sizeof ("s,raw")];
842 	dev_info_t	*dip = NULL;
843 	int		instance, status;
844 	int		num_slices = 1;
845 	int		i;
846 
847 	ASSERT(vdc != NULL);
848 
849 	instance = vdc->instance;
850 	dip = vdc->dip;
851 
852 	switch (vdc->vdisk_type) {
853 	case VD_DISK_TYPE_DISK:
854 		num_slices = V_NUMPAR;
855 		break;
856 	case VD_DISK_TYPE_SLICE:
857 		num_slices = 1;
858 		break;
859 	case VD_DISK_TYPE_UNK:
860 	default:
861 		return (EINVAL);
862 	}
863 
864 	/*
865 	 * Minor nodes are different for EFI disks: EFI disks do not have
866 	 * a minor node 'g' for the minor number corresponding to slice
867 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
868 	 * representing the whole disk.
869 	 */
870 	for (i = 0; i < num_slices; i++) {
871 
872 		if (i == VD_EFI_WD_SLICE) {
873 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
874 				status = vdc_create_device_nodes_efi(vdc);
875 			else
876 				status = vdc_create_device_nodes_vtoc(vdc);
877 			if (status != 0)
878 				return (status);
879 			continue;
880 		}
881 
882 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
883 		if (ddi_create_minor_node(dip, name, S_IFBLK,
884 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
885 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
886 			    instance, name);
887 			return (EIO);
888 		}
889 
890 		/* if any device node is created we set this flag */
891 		vdc->initialized |= VDC_MINOR;
892 
893 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
894 
895 		if (ddi_create_minor_node(dip, name, S_IFCHR,
896 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
897 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
898 			    instance, name);
899 			return (EIO);
900 		}
901 	}
902 
903 	return (0);
904 }
905 
906 /*
907  * Function:
908  *	vdc_create_device_nodes_props
909  *
910  * Description:
911  *	This function creates the block and character device nodes under
912  *	/devices along with the node properties. It is called as part of
913  *	the attach(9E) of the instance during the handshake with vds after
914  *	vds has sent the attributes to vdc.
915  *
916  * Parameters:
917  *	vdc 		- soft state pointer
918  *
919  * Return Values
920  *	0		- Success
921  *	EIO		- Failed to create device node property
922  *	EINVAL		- Unknown type of disk exported
923  */
924 static int
925 vdc_create_device_nodes_props(vdc_t *vdc)
926 {
927 	dev_info_t	*dip = NULL;
928 	int		instance;
929 	int		num_slices = 1;
930 	int64_t		size = 0;
931 	dev_t		dev;
932 	int		rv;
933 	int		i;
934 
935 	ASSERT(vdc != NULL);
936 
937 	instance = vdc->instance;
938 	dip = vdc->dip;
939 
940 	if ((vdc->vtoc == NULL) || (vdc->vtoc->v_sanity != VTOC_SANE)) {
941 		DMSG(vdc, 0, "![%d] Could not create device node property."
942 		    " No VTOC available", instance);
943 		return (ENXIO);
944 	}
945 
946 	switch (vdc->vdisk_type) {
947 	case VD_DISK_TYPE_DISK:
948 		num_slices = V_NUMPAR;
949 		break;
950 	case VD_DISK_TYPE_SLICE:
951 		num_slices = 1;
952 		break;
953 	case VD_DISK_TYPE_UNK:
954 	default:
955 		return (EINVAL);
956 	}
957 
958 	for (i = 0; i < num_slices; i++) {
959 		dev = makedevice(ddi_driver_major(dip),
960 		    VD_MAKE_DEV(instance, i));
961 
962 		size = vdc->vtoc->v_part[i].p_size * vdc->vtoc->v_sectorsz;
963 		DMSG(vdc, 0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
964 		    instance, size, size / (1024 * 1024),
965 		    vdc->vtoc->v_part[i].p_size);
966 
967 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
968 		if (rv != DDI_PROP_SUCCESS) {
969 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
970 			    instance, VDC_SIZE_PROP_NAME, size);
971 			return (EIO);
972 		}
973 
974 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
975 		    lbtodb(size));
976 		if (rv != DDI_PROP_SUCCESS) {
977 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
978 			    instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
979 			return (EIO);
980 		}
981 	}
982 
983 	return (0);
984 }
985 
986 static int
987 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
988 {
989 	_NOTE(ARGUNUSED(cred))
990 
991 	int		instance;
992 	vdc_t		*vdc;
993 
994 	ASSERT(dev != NULL);
995 	instance = VDCUNIT(*dev);
996 
997 	if ((otyp != OTYP_CHR) && (otyp != OTYP_BLK))
998 		return (EINVAL);
999 
1000 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1001 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1002 		return (ENXIO);
1003 	}
1004 
1005 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1006 	    getminor(*dev), flag, otyp);
1007 
1008 	mutex_enter(&vdc->lock);
1009 	vdc->open_count++;
1010 	mutex_exit(&vdc->lock);
1011 
1012 	return (0);
1013 }
1014 
1015 static int
1016 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1017 {
1018 	_NOTE(ARGUNUSED(cred))
1019 
1020 	int	instance;
1021 	vdc_t	*vdc;
1022 
1023 	instance = VDCUNIT(dev);
1024 
1025 	if ((otyp != OTYP_CHR) && (otyp != OTYP_BLK))
1026 		return (EINVAL);
1027 
1028 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1029 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1030 		return (ENXIO);
1031 	}
1032 
1033 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1034 	if (vdc->dkio_flush_pending) {
1035 		DMSG(vdc, 0,
1036 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
1037 		    instance, vdc->dkio_flush_pending);
1038 		return (EBUSY);
1039 	}
1040 
1041 	/*
1042 	 * Should not need the mutex here, since the framework should protect
1043 	 * against more opens on this device, but just in case.
1044 	 */
1045 	mutex_enter(&vdc->lock);
1046 	vdc->open_count--;
1047 	mutex_exit(&vdc->lock);
1048 
1049 	return (0);
1050 }
1051 
1052 static int
1053 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1054 {
1055 	_NOTE(ARGUNUSED(credp))
1056 	_NOTE(ARGUNUSED(rvalp))
1057 
1058 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode));
1059 }
1060 
1061 static int
1062 vdc_print(dev_t dev, char *str)
1063 {
1064 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1065 	return (0);
1066 }
1067 
1068 static int
1069 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1070 {
1071 	int	rv;
1072 	size_t	nbytes = nblk * DEV_BSIZE;
1073 	int	instance = VDCUNIT(dev);
1074 	vdc_t	*vdc = NULL;
1075 
1076 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1077 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1078 		return (ENXIO);
1079 	}
1080 
1081 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1082 	    instance, nbytes, blkno, (void *)addr);
1083 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1084 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1085 	if (rv) {
1086 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1087 		return (rv);
1088 	}
1089 
1090 	if (ddi_in_panic())
1091 		(void) vdc_drain_response(vdc);
1092 
1093 	DMSG(vdc, 0, "[%d] End\n", instance);
1094 
1095 	return (0);
1096 }
1097 
1098 /* -------------------------------------------------------------------------- */
1099 
1100 /*
1101  * Disk access routines
1102  *
1103  */
1104 
1105 /*
1106  * vdc_strategy()
1107  *
1108  * Return Value:
1109  *	0:	As per strategy(9E), the strategy() function must return 0
1110  *		[ bioerror(9f) sets b_flags to the proper error code ]
1111  */
1112 static int
1113 vdc_strategy(struct buf *buf)
1114 {
1115 	int	rv = -1;
1116 	vdc_t	*vdc = NULL;
1117 	int	instance = VDCUNIT(buf->b_edev);
1118 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1119 	int	slice;
1120 
1121 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1122 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1123 		bioerror(buf, ENXIO);
1124 		biodone(buf);
1125 		return (0);
1126 	}
1127 
1128 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1129 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1130 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1131 	DTRACE_IO2(vstart, buf_t *, buf, vdc_t *, vdc);
1132 
1133 	bp_mapin(buf);
1134 
1135 	if ((long)buf->b_private == VD_SLICE_NONE) {
1136 		/* I/O using an absolute disk offset */
1137 		slice = VD_SLICE_NONE;
1138 	} else {
1139 		slice = VDCPART(buf->b_edev);
1140 	}
1141 
1142 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1143 	    buf->b_bcount, slice, buf->b_lblkno,
1144 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1145 	    VIO_write_dir);
1146 
1147 	/*
1148 	 * If the request was successfully sent, the strategy call returns and
1149 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1150 	 * done.
1151 	 */
1152 	if (rv) {
1153 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1154 		bioerror(buf, rv);
1155 		biodone(buf);
1156 	}
1157 
1158 	return (0);
1159 }
1160 
1161 /*
1162  * Function:
1163  *	vdc_min
1164  *
1165  * Description:
1166  *	Routine to limit the size of a data transfer. Used in
1167  *	conjunction with physio(9F).
1168  *
1169  * Arguments:
1170  *	bp - pointer to the indicated buf(9S) struct.
1171  *
1172  */
1173 static void
1174 vdc_min(struct buf *bufp)
1175 {
1176 	vdc_t	*vdc = NULL;
1177 	int	instance = VDCUNIT(bufp->b_edev);
1178 
1179 	vdc = ddi_get_soft_state(vdc_state, instance);
1180 	VERIFY(vdc != NULL);
1181 
1182 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1183 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1184 	}
1185 }
1186 
1187 static int
1188 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1189 {
1190 	_NOTE(ARGUNUSED(cred))
1191 
1192 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1193 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1194 }
1195 
1196 static int
1197 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1198 {
1199 	_NOTE(ARGUNUSED(cred))
1200 
1201 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1202 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1203 }
1204 
1205 static int
1206 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1207 {
1208 	_NOTE(ARGUNUSED(cred))
1209 
1210 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1211 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1212 }
1213 
1214 static int
1215 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1216 {
1217 	_NOTE(ARGUNUSED(cred))
1218 
1219 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1220 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1221 }
1222 
1223 
1224 /* -------------------------------------------------------------------------- */
1225 
1226 /*
1227  * Handshake support
1228  */
1229 
1230 
1231 /*
1232  * Function:
1233  *	vdc_init_ver_negotiation()
1234  *
1235  * Description:
1236  *
1237  * Arguments:
1238  *	vdc	- soft state pointer for this instance of the device driver.
1239  *
1240  * Return Code:
1241  *	0	- Success
1242  */
1243 static int
1244 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1245 {
1246 	vio_ver_msg_t	pkt;
1247 	size_t		msglen = sizeof (pkt);
1248 	int		status = -1;
1249 
1250 	ASSERT(vdc != NULL);
1251 	ASSERT(mutex_owned(&vdc->lock));
1252 
1253 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1254 
1255 	/*
1256 	 * set the Session ID to a unique value
1257 	 * (the lower 32 bits of the clock tick)
1258 	 */
1259 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1260 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1261 
1262 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1263 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1264 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1265 	pkt.tag.vio_sid = vdc->session_id;
1266 	pkt.dev_class = VDEV_DISK;
1267 	pkt.ver_major = ver.major;
1268 	pkt.ver_minor = ver.minor;
1269 
1270 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1271 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1272 	    vdc->instance, status);
1273 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1274 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1275 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1276 		    status, msglen);
1277 		if (msglen != sizeof (vio_ver_msg_t))
1278 			status = ENOMSG;
1279 	}
1280 
1281 	return (status);
1282 }
1283 
1284 /*
1285  * Function:
1286  *	vdc_ver_negotiation()
1287  *
1288  * Description:
1289  *
1290  * Arguments:
1291  *	vdcp	- soft state pointer for this instance of the device driver.
1292  *
1293  * Return Code:
1294  *	0	- Success
1295  */
1296 static int
1297 vdc_ver_negotiation(vdc_t *vdcp)
1298 {
1299 	vio_msg_t vio_msg;
1300 	int status;
1301 
1302 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1303 		return (status);
1304 
1305 	/* release lock and wait for response */
1306 	mutex_exit(&vdcp->lock);
1307 	status = vdc_wait_for_response(vdcp, &vio_msg);
1308 	mutex_enter(&vdcp->lock);
1309 	if (status) {
1310 		DMSG(vdcp, 0,
1311 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1312 		    vdcp->instance, status);
1313 		return (status);
1314 	}
1315 
1316 	/* check type and sub_type ... */
1317 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1318 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1319 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1320 		    vdcp->instance);
1321 		return (EPROTO);
1322 	}
1323 
1324 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1325 }
1326 
1327 /*
1328  * Function:
1329  *	vdc_init_attr_negotiation()
1330  *
1331  * Description:
1332  *
1333  * Arguments:
1334  *	vdc	- soft state pointer for this instance of the device driver.
1335  *
1336  * Return Code:
1337  *	0	- Success
1338  */
1339 static int
1340 vdc_init_attr_negotiation(vdc_t *vdc)
1341 {
1342 	vd_attr_msg_t	pkt;
1343 	size_t		msglen = sizeof (pkt);
1344 	int		status;
1345 
1346 	ASSERT(vdc != NULL);
1347 	ASSERT(mutex_owned(&vdc->lock));
1348 
1349 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1350 
1351 	/* fill in tag */
1352 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1353 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1354 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1355 	pkt.tag.vio_sid = vdc->session_id;
1356 	/* fill in payload */
1357 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1358 	pkt.vdisk_block_size = vdc->block_size;
1359 	pkt.xfer_mode = VIO_DRING_MODE;
1360 	pkt.operations = 0;	/* server will set bits of valid operations */
1361 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1362 	pkt.vdisk_size = 0;	/* server will set to valid size */
1363 
1364 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1365 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1366 
1367 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1368 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1369 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1370 		    status, msglen);
1371 		if (msglen != sizeof (vio_ver_msg_t))
1372 			status = ENOMSG;
1373 	}
1374 
1375 	return (status);
1376 }
1377 
1378 /*
1379  * Function:
1380  *	vdc_attr_negotiation()
1381  *
1382  * Description:
1383  *
1384  * Arguments:
1385  *	vdc	- soft state pointer for this instance of the device driver.
1386  *
1387  * Return Code:
1388  *	0	- Success
1389  */
1390 static int
1391 vdc_attr_negotiation(vdc_t *vdcp)
1392 {
1393 	int status;
1394 	vio_msg_t vio_msg;
1395 
1396 	if (status = vdc_init_attr_negotiation(vdcp))
1397 		return (status);
1398 
1399 	/* release lock and wait for response */
1400 	mutex_exit(&vdcp->lock);
1401 	status = vdc_wait_for_response(vdcp, &vio_msg);
1402 	mutex_enter(&vdcp->lock);
1403 	if (status) {
1404 		DMSG(vdcp, 0,
1405 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1406 		    vdcp->instance, status);
1407 		return (status);
1408 	}
1409 
1410 	/* check type and sub_type ... */
1411 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1412 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1413 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1414 		    vdcp->instance);
1415 		return (EPROTO);
1416 	}
1417 
1418 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1419 }
1420 
1421 
1422 /*
1423  * Function:
1424  *	vdc_init_dring_negotiate()
1425  *
1426  * Description:
1427  *
1428  * Arguments:
1429  *	vdc	- soft state pointer for this instance of the device driver.
1430  *
1431  * Return Code:
1432  *	0	- Success
1433  */
1434 static int
1435 vdc_init_dring_negotiate(vdc_t *vdc)
1436 {
1437 	vio_dring_reg_msg_t	pkt;
1438 	size_t			msglen = sizeof (pkt);
1439 	int			status = -1;
1440 	int			retry;
1441 	int			nretries = 10;
1442 
1443 	ASSERT(vdc != NULL);
1444 	ASSERT(mutex_owned(&vdc->lock));
1445 
1446 	for (retry = 0; retry < nretries; retry++) {
1447 		status = vdc_init_descriptor_ring(vdc);
1448 		if (status != EAGAIN)
1449 			break;
1450 		drv_usecwait(vdc_min_timeout_ldc);
1451 	}
1452 
1453 	if (status != 0) {
1454 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1455 		    vdc->instance, status);
1456 		return (status);
1457 	}
1458 
1459 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1460 	    vdc->instance, status);
1461 
1462 	/* fill in tag */
1463 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1464 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1465 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1466 	pkt.tag.vio_sid = vdc->session_id;
1467 	/* fill in payload */
1468 	pkt.dring_ident = 0;
1469 	pkt.num_descriptors = vdc->dring_len;
1470 	pkt.descriptor_size = vdc->dring_entry_size;
1471 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1472 	pkt.ncookies = vdc->dring_cookie_count;
1473 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1474 
1475 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1476 	if (status != 0) {
1477 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1478 		    vdc->instance, status);
1479 	}
1480 
1481 	return (status);
1482 }
1483 
1484 
1485 /*
1486  * Function:
1487  *	vdc_dring_negotiation()
1488  *
1489  * Description:
1490  *
1491  * Arguments:
1492  *	vdc	- soft state pointer for this instance of the device driver.
1493  *
1494  * Return Code:
1495  *	0	- Success
1496  */
1497 static int
1498 vdc_dring_negotiation(vdc_t *vdcp)
1499 {
1500 	int status;
1501 	vio_msg_t vio_msg;
1502 
1503 	if (status = vdc_init_dring_negotiate(vdcp))
1504 		return (status);
1505 
1506 	/* release lock and wait for response */
1507 	mutex_exit(&vdcp->lock);
1508 	status = vdc_wait_for_response(vdcp, &vio_msg);
1509 	mutex_enter(&vdcp->lock);
1510 	if (status) {
1511 		DMSG(vdcp, 0,
1512 		    "[%d] Failed waiting for Dring negotiation response,"
1513 		    " rv(%d)", vdcp->instance, status);
1514 		return (status);
1515 	}
1516 
1517 	/* check type and sub_type ... */
1518 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1519 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1520 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1521 		    vdcp->instance);
1522 		return (EPROTO);
1523 	}
1524 
1525 	return (vdc_handle_dring_reg_msg(vdcp,
1526 	    (vio_dring_reg_msg_t *)&vio_msg));
1527 }
1528 
1529 
1530 /*
1531  * Function:
1532  *	vdc_send_rdx()
1533  *
1534  * Description:
1535  *
1536  * Arguments:
1537  *	vdc	- soft state pointer for this instance of the device driver.
1538  *
1539  * Return Code:
1540  *	0	- Success
1541  */
1542 static int
1543 vdc_send_rdx(vdc_t *vdcp)
1544 {
1545 	vio_msg_t	msg;
1546 	size_t		msglen = sizeof (vio_msg_t);
1547 	int		status;
1548 
1549 	/*
1550 	 * Send an RDX message to vds to indicate we are ready
1551 	 * to send data
1552 	 */
1553 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1554 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1555 	msg.tag.vio_subtype_env = VIO_RDX;
1556 	msg.tag.vio_sid = vdcp->session_id;
1557 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1558 	if (status != 0) {
1559 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1560 		    vdcp->instance, status);
1561 	}
1562 
1563 	return (status);
1564 }
1565 
1566 /*
1567  * Function:
1568  *	vdc_handle_rdx()
1569  *
1570  * Description:
1571  *
1572  * Arguments:
1573  *	vdc	- soft state pointer for this instance of the device driver.
1574  *	msgp	- received msg
1575  *
1576  * Return Code:
1577  *	0	- Success
1578  */
1579 static int
1580 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1581 {
1582 	_NOTE(ARGUNUSED(vdcp))
1583 	_NOTE(ARGUNUSED(msgp))
1584 
1585 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1586 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1587 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1588 
1589 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1590 
1591 	return (0);
1592 }
1593 
1594 /*
1595  * Function:
1596  *	vdc_rdx_exchange()
1597  *
1598  * Description:
1599  *
1600  * Arguments:
1601  *	vdc	- soft state pointer for this instance of the device driver.
1602  *
1603  * Return Code:
1604  *	0	- Success
1605  */
1606 static int
1607 vdc_rdx_exchange(vdc_t *vdcp)
1608 {
1609 	int status;
1610 	vio_msg_t vio_msg;
1611 
1612 	if (status = vdc_send_rdx(vdcp))
1613 		return (status);
1614 
1615 	/* release lock and wait for response */
1616 	mutex_exit(&vdcp->lock);
1617 	status = vdc_wait_for_response(vdcp, &vio_msg);
1618 	mutex_enter(&vdcp->lock);
1619 	if (status) {
1620 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1621 		    vdcp->instance, status);
1622 		return (status);
1623 	}
1624 
1625 	/* check type and sub_type ... */
1626 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1627 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1628 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1629 		return (EPROTO);
1630 	}
1631 
1632 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1633 }
1634 
1635 
1636 /* -------------------------------------------------------------------------- */
1637 
1638 /*
1639  * LDC helper routines
1640  */
1641 
1642 static int
1643 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1644 {
1645 	int		status;
1646 	boolean_t	q_has_pkts = B_FALSE;
1647 	int		delay_time;
1648 	size_t		len;
1649 
1650 	mutex_enter(&vdc->read_lock);
1651 
1652 	if (vdc->read_state == VDC_READ_IDLE)
1653 		vdc->read_state = VDC_READ_WAITING;
1654 
1655 	while (vdc->read_state != VDC_READ_PENDING) {
1656 
1657 		/* detect if the connection has been reset */
1658 		if (vdc->read_state == VDC_READ_RESET) {
1659 			status = ECONNRESET;
1660 			goto done;
1661 		}
1662 
1663 		cv_wait(&vdc->read_cv, &vdc->read_lock);
1664 	}
1665 
1666 	/*
1667 	 * Until we get a blocking ldc read we have to retry
1668 	 * until the entire LDC message has arrived before
1669 	 * ldc_read() will succeed. Note we also bail out if
1670 	 * the channel is reset or goes away.
1671 	 */
1672 	delay_time = vdc_ldc_read_init_delay;
1673 loop:
1674 	len = *nbytesp;
1675 	status = ldc_read(vdc->ldc_handle, (caddr_t)msgp, &len);
1676 	switch (status) {
1677 	case EAGAIN:
1678 		delay_time *= 2;
1679 		if (delay_time >= vdc_ldc_read_max_delay)
1680 			delay_time = vdc_ldc_read_max_delay;
1681 		delay(delay_time);
1682 		goto loop;
1683 
1684 	case 0:
1685 		if (len == 0) {
1686 			DMSG(vdc, 0, "[%d] ldc_read returned 0 bytes with "
1687 			    "no error!\n", vdc->instance);
1688 			goto loop;
1689 		}
1690 
1691 		*nbytesp = len;
1692 
1693 		/*
1694 		 * If there are pending messages, leave the
1695 		 * read state as pending. Otherwise, set the state
1696 		 * back to idle.
1697 		 */
1698 		status = ldc_chkq(vdc->ldc_handle, &q_has_pkts);
1699 		if (status == 0 && !q_has_pkts)
1700 			vdc->read_state = VDC_READ_IDLE;
1701 
1702 		break;
1703 	default:
1704 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
1705 		break;
1706 	}
1707 
1708 done:
1709 	mutex_exit(&vdc->read_lock);
1710 
1711 	return (status);
1712 }
1713 
1714 
1715 
1716 #ifdef DEBUG
1717 void
1718 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
1719 {
1720 	char *ms, *ss, *ses;
1721 	switch (msg->tag.vio_msgtype) {
1722 #define	Q(_s)	case _s : ms = #_s; break;
1723 	Q(VIO_TYPE_CTRL)
1724 	Q(VIO_TYPE_DATA)
1725 	Q(VIO_TYPE_ERR)
1726 #undef Q
1727 	default: ms = "unknown"; break;
1728 	}
1729 
1730 	switch (msg->tag.vio_subtype) {
1731 #define	Q(_s)	case _s : ss = #_s; break;
1732 	Q(VIO_SUBTYPE_INFO)
1733 	Q(VIO_SUBTYPE_ACK)
1734 	Q(VIO_SUBTYPE_NACK)
1735 #undef Q
1736 	default: ss = "unknown"; break;
1737 	}
1738 
1739 	switch (msg->tag.vio_subtype_env) {
1740 #define	Q(_s)	case _s : ses = #_s; break;
1741 	Q(VIO_VER_INFO)
1742 	Q(VIO_ATTR_INFO)
1743 	Q(VIO_DRING_REG)
1744 	Q(VIO_DRING_UNREG)
1745 	Q(VIO_RDX)
1746 	Q(VIO_PKT_DATA)
1747 	Q(VIO_DESC_DATA)
1748 	Q(VIO_DRING_DATA)
1749 #undef Q
1750 	default: ses = "unknown"; break;
1751 	}
1752 
1753 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
1754 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
1755 	    msg->tag.vio_subtype_env, ms, ss, ses);
1756 }
1757 #endif
1758 
1759 /*
1760  * Function:
1761  *	vdc_send()
1762  *
1763  * Description:
1764  *	The function encapsulates the call to write a message using LDC.
1765  *	If LDC indicates that the call failed due to the queue being full,
1766  *	we retry the ldc_write() [ up to 'vdc_retries' time ], otherwise
1767  *	we return the error returned by LDC.
1768  *
1769  * Arguments:
1770  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
1771  *	pkt		- address of LDC message to be sent
1772  *	msglen		- the size of the message being sent. When the function
1773  *			  returns, this contains the number of bytes written.
1774  *
1775  * Return Code:
1776  *	0		- Success.
1777  *	EINVAL		- pkt or msglen were NULL
1778  *	ECONNRESET	- The connection was not up.
1779  *	EWOULDBLOCK	- LDC queue is full
1780  *	xxx		- other error codes returned by ldc_write
1781  */
1782 static int
1783 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
1784 {
1785 	size_t	size = 0;
1786 	int	status = 0;
1787 	clock_t delay_ticks;
1788 
1789 	ASSERT(vdc != NULL);
1790 	ASSERT(mutex_owned(&vdc->lock));
1791 	ASSERT(msglen != NULL);
1792 	ASSERT(*msglen != 0);
1793 
1794 #ifdef DEBUG
1795 	vdc_decode_tag(vdc, (vio_msg_t *)pkt);
1796 #endif
1797 	/*
1798 	 * Wait indefinitely to send if channel
1799 	 * is busy, but bail out if we succeed or
1800 	 * if the channel closes or is reset.
1801 	 */
1802 	delay_ticks = vdc_hz_min_ldc_delay;
1803 	do {
1804 		size = *msglen;
1805 		status = ldc_write(vdc->ldc_handle, pkt, &size);
1806 		if (status == EWOULDBLOCK) {
1807 			delay(delay_ticks);
1808 			/* geometric backoff */
1809 			delay_ticks *= 2;
1810 			if (delay_ticks > vdc_hz_max_ldc_delay)
1811 				delay_ticks = vdc_hz_max_ldc_delay;
1812 		}
1813 	} while (status == EWOULDBLOCK);
1814 
1815 	/* if LDC had serious issues --- reset vdc state */
1816 	if (status == EIO || status == ECONNRESET) {
1817 		/* LDC had serious issues --- reset vdc state */
1818 		mutex_enter(&vdc->read_lock);
1819 		if ((vdc->read_state == VDC_READ_WAITING) ||
1820 		    (vdc->read_state == VDC_READ_RESET))
1821 			cv_signal(&vdc->read_cv);
1822 		vdc->read_state = VDC_READ_RESET;
1823 		mutex_exit(&vdc->read_lock);
1824 
1825 		/* wake up any waiters in the reset thread */
1826 		if (vdc->state == VDC_STATE_INIT_WAITING) {
1827 			DMSG(vdc, 0, "[%d] write reset - "
1828 			    "vdc is resetting ..\n", vdc->instance);
1829 			vdc->state = VDC_STATE_RESETTING;
1830 			cv_signal(&vdc->initwait_cv);
1831 		}
1832 
1833 		return (ECONNRESET);
1834 	}
1835 
1836 	/* return the last size written */
1837 	*msglen = size;
1838 
1839 	return (status);
1840 }
1841 
1842 /*
1843  * Function:
1844  *	vdc_get_md_node
1845  *
1846  * Description:
1847  *	Get the MD, the device node and the port node for the given
1848  *	disk instance. The caller is responsible for cleaning up the
1849  *	reference to the returned MD (mdpp) by calling md_fini_handle().
1850  *
1851  * Arguments:
1852  *	dip	- dev info pointer for this instance of the device driver.
1853  *	mdpp	- the returned MD.
1854  *	vd_nodep - the returned device node.
1855  *	vd_portp - the returned port node. The returned port node is NULL
1856  *		   if no port node is found.
1857  *
1858  * Return Code:
1859  *	0	- Success.
1860  *	ENOENT	- Expected node or property did not exist.
1861  *	ENXIO	- Unexpected error communicating with MD framework
1862  */
1863 static int
1864 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep,
1865     mde_cookie_t *vd_portp)
1866 {
1867 	int		status = ENOENT;
1868 	char		*node_name = NULL;
1869 	md_t		*mdp = NULL;
1870 	int		num_nodes;
1871 	int		num_vdevs;
1872 	int		num_vports;
1873 	mde_cookie_t	rootnode;
1874 	mde_cookie_t	*listp = NULL;
1875 	boolean_t	found_inst = B_FALSE;
1876 	int		listsz;
1877 	int		idx;
1878 	uint64_t	md_inst;
1879 	int		obp_inst;
1880 	int		instance = ddi_get_instance(dip);
1881 
1882 	/*
1883 	 * Get the OBP instance number for comparison with the MD instance
1884 	 *
1885 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
1886 	 * notion of "instance", or unique identifier, for that node; OBP
1887 	 * stores the value of the "cfg-handle" MD property as the value of
1888 	 * the "reg" property on the node in the device tree it builds from
1889 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
1890 	 * "reg" property value to uniquely identify this device instance.
1891 	 * If the "reg" property cannot be found, the device tree state is
1892 	 * presumably so broken that there is no point in continuing.
1893 	 */
1894 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
1895 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
1896 		return (ENOENT);
1897 	}
1898 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
1899 	    OBP_REG, -1);
1900 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
1901 
1902 	/*
1903 	 * We now walk the MD nodes to find the node for this vdisk.
1904 	 */
1905 	if ((mdp = md_get_handle()) == NULL) {
1906 		cmn_err(CE_WARN, "unable to init machine description");
1907 		return (ENXIO);
1908 	}
1909 
1910 	num_nodes = md_node_count(mdp);
1911 	ASSERT(num_nodes > 0);
1912 
1913 	listsz = num_nodes * sizeof (mde_cookie_t);
1914 
1915 	/* allocate memory for nodes */
1916 	listp = kmem_zalloc(listsz, KM_SLEEP);
1917 
1918 	rootnode = md_root_node(mdp);
1919 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
1920 
1921 	/*
1922 	 * Search for all the virtual devices, we will then check to see which
1923 	 * ones are disk nodes.
1924 	 */
1925 	num_vdevs = md_scan_dag(mdp, rootnode,
1926 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
1927 	    md_find_name(mdp, "fwd"), listp);
1928 
1929 	if (num_vdevs <= 0) {
1930 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
1931 		status = ENOENT;
1932 		goto done;
1933 	}
1934 
1935 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
1936 	for (idx = 0; idx < num_vdevs; idx++) {
1937 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
1938 		if ((status != 0) || (node_name == NULL)) {
1939 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
1940 			    ": err %d", VDC_MD_VDEV_NAME, status);
1941 			continue;
1942 		}
1943 
1944 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
1945 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
1946 			status = md_get_prop_val(mdp, listp[idx],
1947 			    VDC_MD_CFG_HDL, &md_inst);
1948 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
1949 			    instance, md_inst);
1950 			if ((status == 0) && (md_inst == obp_inst)) {
1951 				found_inst = B_TRUE;
1952 				break;
1953 			}
1954 		}
1955 	}
1956 
1957 	if (!found_inst) {
1958 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
1959 		status = ENOENT;
1960 		goto done;
1961 	}
1962 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
1963 
1964 	*vd_nodep = listp[idx];
1965 	*mdpp = mdp;
1966 
1967 	num_vports = md_scan_dag(mdp, *vd_nodep,
1968 	    md_find_name(mdp, VDC_MD_PORT_NAME),
1969 	    md_find_name(mdp, "fwd"), listp);
1970 
1971 	if (num_vports != 1) {
1972 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
1973 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME, num_vports);
1974 	}
1975 
1976 	*vd_portp = (num_vports == 0)? NULL: listp[0];
1977 
1978 done:
1979 	kmem_free(listp, listsz);
1980 	return (status);
1981 }
1982 
1983 /*
1984  * Function:
1985  *	vdc_get_ldc_id()
1986  *
1987  * Description:
1988  *	This function gets the 'ldc-id' for this particular instance of vdc.
1989  *	The id returned is the guest domain channel endpoint LDC uses for
1990  *	communication with vds.
1991  *
1992  * Arguments:
1993  *	mdp	- pointer to the machine description.
1994  *	vd_node	- the vdisk element from the MD.
1995  *	ldc_id	- pointer to variable used to return the 'ldc-id' found.
1996  *
1997  * Return Code:
1998  *	0	- Success.
1999  *	ENOENT	- Expected node or property did not exist.
2000  */
2001 static int
2002 vdc_get_ldc_id(md_t *mdp, mde_cookie_t vd_node, uint64_t *ldc_id)
2003 {
2004 	mde_cookie_t	*chanp = NULL;
2005 	int		listsz;
2006 	int		num_chans;
2007 	int		num_nodes;
2008 	int		status = 0;
2009 
2010 	num_nodes = md_node_count(mdp);
2011 	ASSERT(num_nodes > 0);
2012 
2013 	listsz = num_nodes * sizeof (mde_cookie_t);
2014 
2015 	/* allocate memory for nodes */
2016 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2017 
2018 	/* get the channels for this node */
2019 	num_chans = md_scan_dag(mdp, vd_node,
2020 	    md_find_name(mdp, VDC_MD_CHAN_NAME),
2021 	    md_find_name(mdp, "fwd"), chanp);
2022 
2023 	/* expecting at least one channel */
2024 	if (num_chans <= 0) {
2025 		cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2026 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2027 		status = ENOENT;
2028 		goto done;
2029 
2030 	} else if (num_chans != 1) {
2031 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2032 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME, num_chans);
2033 	}
2034 
2035 	/*
2036 	 * We use the first channel found (index 0), irrespective of how
2037 	 * many are there in total.
2038 	 */
2039 	if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID, ldc_id) != 0) {
2040 		cmn_err(CE_NOTE, "Channel '%s' property not found", VDC_MD_ID);
2041 		status = ENOENT;
2042 	}
2043 
2044 done:
2045 	kmem_free(chanp, listsz);
2046 	return (status);
2047 }
2048 
2049 static int
2050 vdc_do_ldc_up(vdc_t *vdc)
2051 {
2052 	int		status;
2053 	ldc_status_t	ldc_state;
2054 
2055 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2056 	    vdc->instance, vdc->ldc_id);
2057 
2058 	if (vdc->lifecycle == VDC_LC_DETACHING)
2059 		return (EINVAL);
2060 
2061 	if ((status = ldc_up(vdc->ldc_handle)) != 0) {
2062 		switch (status) {
2063 		case ECONNREFUSED:	/* listener not ready at other end */
2064 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2065 			    vdc->instance, vdc->ldc_id, status);
2066 			status = 0;
2067 			break;
2068 		default:
2069 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2070 			    "channel=%ld, err=%d", vdc->instance, vdc->ldc_id,
2071 			    status);
2072 			break;
2073 		}
2074 	}
2075 
2076 	if (ldc_status(vdc->ldc_handle, &ldc_state) == 0) {
2077 		vdc->ldc_state = ldc_state;
2078 		if (ldc_state == LDC_UP) {
2079 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2080 			    vdc->instance);
2081 			vdc->seq_num = 1;
2082 			vdc->seq_num_reply = 0;
2083 		}
2084 	}
2085 
2086 	return (status);
2087 }
2088 
2089 /*
2090  * Function:
2091  *	vdc_terminate_ldc()
2092  *
2093  * Description:
2094  *
2095  * Arguments:
2096  *	vdc	- soft state pointer for this instance of the device driver.
2097  *
2098  * Return Code:
2099  *	None
2100  */
2101 static void
2102 vdc_terminate_ldc(vdc_t *vdc)
2103 {
2104 	int	instance = ddi_get_instance(vdc->dip);
2105 
2106 	ASSERT(vdc != NULL);
2107 	ASSERT(mutex_owned(&vdc->lock));
2108 
2109 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2110 
2111 	if (vdc->initialized & VDC_LDC_OPEN) {
2112 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2113 		(void) ldc_close(vdc->ldc_handle);
2114 	}
2115 	if (vdc->initialized & VDC_LDC_CB) {
2116 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2117 		(void) ldc_unreg_callback(vdc->ldc_handle);
2118 	}
2119 	if (vdc->initialized & VDC_LDC) {
2120 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2121 		(void) ldc_fini(vdc->ldc_handle);
2122 		vdc->ldc_handle = NULL;
2123 	}
2124 
2125 	vdc->initialized &= ~(VDC_LDC | VDC_LDC_CB | VDC_LDC_OPEN);
2126 }
2127 
2128 /* -------------------------------------------------------------------------- */
2129 
2130 /*
2131  * Descriptor Ring helper routines
2132  */
2133 
2134 /*
2135  * Function:
2136  *	vdc_init_descriptor_ring()
2137  *
2138  * Description:
2139  *
2140  * Arguments:
2141  *	vdc	- soft state pointer for this instance of the device driver.
2142  *
2143  * Return Code:
2144  *	0	- Success
2145  */
2146 static int
2147 vdc_init_descriptor_ring(vdc_t *vdc)
2148 {
2149 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2150 	int	status = 0;
2151 	int	i;
2152 
2153 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2154 
2155 	ASSERT(vdc != NULL);
2156 	ASSERT(mutex_owned(&vdc->lock));
2157 	ASSERT(vdc->ldc_handle != NULL);
2158 
2159 	/* ensure we have enough room to store max sized block */
2160 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2161 
2162 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2163 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2164 		/*
2165 		 * Calculate the maximum block size we can transmit using one
2166 		 * Descriptor Ring entry from the attributes returned by the
2167 		 * vDisk server. This is subject to a minimum of 'maxphys'
2168 		 * as we do not have the capability to split requests over
2169 		 * multiple DRing entries.
2170 		 */
2171 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2172 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2173 			    vdc->instance);
2174 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2175 		} else {
2176 			vdc->dring_max_cookies =
2177 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2178 		}
2179 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2180 		    (sizeof (ldc_mem_cookie_t) *
2181 		    (vdc->dring_max_cookies - 1)));
2182 		vdc->dring_len = VD_DRING_LEN;
2183 
2184 		status = ldc_mem_dring_create(vdc->dring_len,
2185 		    vdc->dring_entry_size, &vdc->ldc_dring_hdl);
2186 		if ((vdc->ldc_dring_hdl == NULL) || (status != 0)) {
2187 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2188 			    vdc->instance);
2189 			return (status);
2190 		}
2191 		vdc->initialized |= VDC_DRING_INIT;
2192 	}
2193 
2194 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2195 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2196 		vdc->dring_cookie =
2197 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2198 
2199 		status = ldc_mem_dring_bind(vdc->ldc_handle, vdc->ldc_dring_hdl,
2200 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2201 		    &vdc->dring_cookie[0],
2202 		    &vdc->dring_cookie_count);
2203 		if (status != 0) {
2204 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2205 			    "(%lx) to channel (%lx) status=%d\n",
2206 			    vdc->instance, vdc->ldc_dring_hdl,
2207 			    vdc->ldc_handle, status);
2208 			return (status);
2209 		}
2210 		ASSERT(vdc->dring_cookie_count == 1);
2211 		vdc->initialized |= VDC_DRING_BOUND;
2212 	}
2213 
2214 	status = ldc_mem_dring_info(vdc->ldc_dring_hdl, &vdc->dring_mem_info);
2215 	if (status != 0) {
2216 		DMSG(vdc, 0,
2217 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2218 		    vdc->instance, vdc->ldc_dring_hdl);
2219 		return (status);
2220 	}
2221 
2222 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2223 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2224 
2225 		/* Allocate the local copy of this dring */
2226 		vdc->local_dring =
2227 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2228 		    KM_SLEEP);
2229 		vdc->initialized |= VDC_DRING_LOCAL;
2230 	}
2231 
2232 	/*
2233 	 * Mark all DRing entries as free and initialize the private
2234 	 * descriptor's memory handles. If any entry is initialized,
2235 	 * we need to free it later so we set the bit in 'initialized'
2236 	 * at the start.
2237 	 */
2238 	vdc->initialized |= VDC_DRING_ENTRY;
2239 	for (i = 0; i < vdc->dring_len; i++) {
2240 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2241 		dep->hdr.dstate = VIO_DESC_FREE;
2242 
2243 		status = ldc_mem_alloc_handle(vdc->ldc_handle,
2244 		    &vdc->local_dring[i].desc_mhdl);
2245 		if (status != 0) {
2246 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2247 			    " descriptor %d", vdc->instance, i);
2248 			return (status);
2249 		}
2250 		vdc->local_dring[i].is_free = B_TRUE;
2251 		vdc->local_dring[i].dep = dep;
2252 	}
2253 
2254 	/* Initialize the starting index */
2255 	vdc->dring_curr_idx = 0;
2256 
2257 	return (status);
2258 }
2259 
2260 /*
2261  * Function:
2262  *	vdc_destroy_descriptor_ring()
2263  *
2264  * Description:
2265  *
2266  * Arguments:
2267  *	vdc	- soft state pointer for this instance of the device driver.
2268  *
2269  * Return Code:
2270  *	None
2271  */
2272 static void
2273 vdc_destroy_descriptor_ring(vdc_t *vdc)
2274 {
2275 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2276 	ldc_mem_handle_t	mhdl = NULL;
2277 	ldc_mem_info_t		minfo;
2278 	int			status = -1;
2279 	int			i;	/* loop */
2280 
2281 	ASSERT(vdc != NULL);
2282 	ASSERT(mutex_owned(&vdc->lock));
2283 
2284 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2285 
2286 	if (vdc->initialized & VDC_DRING_ENTRY) {
2287 		DMSG(vdc, 0,
2288 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2289 		for (i = 0; i < vdc->dring_len; i++) {
2290 			ldep = &vdc->local_dring[i];
2291 			mhdl = ldep->desc_mhdl;
2292 
2293 			if (mhdl == NULL)
2294 				continue;
2295 
2296 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2297 				DMSG(vdc, 0,
2298 				    "ldc_mem_info returned an error: %d\n",
2299 				    status);
2300 
2301 				/*
2302 				 * This must mean that the mem handle
2303 				 * is not valid. Clear it out so that
2304 				 * no one tries to use it.
2305 				 */
2306 				ldep->desc_mhdl = NULL;
2307 				continue;
2308 			}
2309 
2310 			if (minfo.status == LDC_BOUND) {
2311 				(void) ldc_mem_unbind_handle(mhdl);
2312 			}
2313 
2314 			(void) ldc_mem_free_handle(mhdl);
2315 
2316 			ldep->desc_mhdl = NULL;
2317 		}
2318 		vdc->initialized &= ~VDC_DRING_ENTRY;
2319 	}
2320 
2321 	if (vdc->initialized & VDC_DRING_LOCAL) {
2322 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2323 		kmem_free(vdc->local_dring,
2324 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2325 		vdc->initialized &= ~VDC_DRING_LOCAL;
2326 	}
2327 
2328 	if (vdc->initialized & VDC_DRING_BOUND) {
2329 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2330 		status = ldc_mem_dring_unbind(vdc->ldc_dring_hdl);
2331 		if (status == 0) {
2332 			vdc->initialized &= ~VDC_DRING_BOUND;
2333 		} else {
2334 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2335 			    vdc->instance, status, vdc->ldc_dring_hdl);
2336 		}
2337 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2338 	}
2339 
2340 	if (vdc->initialized & VDC_DRING_INIT) {
2341 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2342 		status = ldc_mem_dring_destroy(vdc->ldc_dring_hdl);
2343 		if (status == 0) {
2344 			vdc->ldc_dring_hdl = NULL;
2345 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2346 			vdc->initialized &= ~VDC_DRING_INIT;
2347 		} else {
2348 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2349 			    vdc->instance, status, vdc->ldc_dring_hdl);
2350 		}
2351 	}
2352 }
2353 
2354 /*
2355  * Function:
2356  *	vdc_map_to_shared_ring()
2357  *
2358  * Description:
2359  *	Copy contents of the local descriptor to the shared
2360  *	memory descriptor.
2361  *
2362  * Arguments:
2363  *	vdcp	- soft state pointer for this instance of the device driver.
2364  *	idx	- descriptor ring index
2365  *
2366  * Return Code:
2367  *	None
2368  */
2369 static int
2370 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2371 {
2372 	vdc_local_desc_t	*ldep;
2373 	vd_dring_entry_t	*dep;
2374 	int			rv;
2375 
2376 	ldep = &(vdcp->local_dring[idx]);
2377 
2378 	/* for now leave in the old pop_mem_hdl stuff */
2379 	if (ldep->nbytes > 0) {
2380 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2381 		if (rv) {
2382 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2383 			    vdcp->instance);
2384 			return (rv);
2385 		}
2386 	}
2387 
2388 	/*
2389 	 * fill in the data details into the DRing
2390 	 */
2391 	dep = ldep->dep;
2392 	ASSERT(dep != NULL);
2393 
2394 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2395 	dep->payload.operation = ldep->operation;
2396 	dep->payload.addr = ldep->offset;
2397 	dep->payload.nbytes = ldep->nbytes;
2398 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2399 	dep->payload.slice = ldep->slice;
2400 	dep->hdr.dstate = VIO_DESC_READY;
2401 	dep->hdr.ack = 1;		/* request an ACK for every message */
2402 
2403 	return (0);
2404 }
2405 
2406 /*
2407  * Function:
2408  *	vdc_send_request
2409  *
2410  * Description:
2411  *	This routine writes the data to be transmitted to vds into the
2412  *	descriptor, notifies vds that the ring has been updated and
2413  *	then waits for the request to be processed.
2414  *
2415  * Arguments:
2416  *	vdcp	  - the soft state pointer
2417  *	operation - operation we want vds to perform (VD_OP_XXX)
2418  *	addr	  - address of data buf to be read/written.
2419  *	nbytes	  - number of bytes to read/write
2420  *	slice	  - the disk slice this request is for
2421  *	offset	  - relative disk offset
2422  *	cb_type   - type of call - STRATEGY or SYNC
2423  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2424  *			. mode for ioctl(9e)
2425  *			. LP64 diskaddr_t (block I/O)
2426  *	dir	  - direction of operation (READ/WRITE/BOTH)
2427  *
2428  * Return Codes:
2429  *	0
2430  *	ENXIO
2431  */
2432 static int
2433 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2434     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2435     void *cb_arg, vio_desc_direction_t dir)
2436 {
2437 	ASSERT(vdcp != NULL);
2438 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2439 
2440 	mutex_enter(&vdcp->lock);
2441 
2442 	do {
2443 		while (vdcp->state != VDC_STATE_RUNNING) {
2444 
2445 			/* return error if detaching */
2446 			if (vdcp->state == VDC_STATE_DETACH) {
2447 				mutex_exit(&vdcp->lock);
2448 				return (ENXIO);
2449 			}
2450 
2451 			/* fail request if connection timeout is reached */
2452 			if (vdcp->ctimeout_reached) {
2453 				mutex_exit(&vdcp->lock);
2454 				return (EIO);
2455 			}
2456 
2457 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2458 		}
2459 
2460 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2461 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2462 
2463 	mutex_exit(&vdcp->lock);
2464 	return (0);
2465 }
2466 
2467 
2468 /*
2469  * Function:
2470  *	vdc_populate_descriptor
2471  *
2472  * Description:
2473  *	This routine writes the data to be transmitted to vds into the
2474  *	descriptor, notifies vds that the ring has been updated and
2475  *	then waits for the request to be processed.
2476  *
2477  * Arguments:
2478  *	vdcp	  - the soft state pointer
2479  *	operation - operation we want vds to perform (VD_OP_XXX)
2480  *	addr	  - address of data buf to be read/written.
2481  *	nbytes	  - number of bytes to read/write
2482  *	slice	  - the disk slice this request is for
2483  *	offset	  - relative disk offset
2484  *	cb_type   - type of call - STRATEGY or SYNC
2485  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2486  *			. mode for ioctl(9e)
2487  *			. LP64 diskaddr_t (block I/O)
2488  *	dir	  - direction of operation (READ/WRITE/BOTH)
2489  *
2490  * Return Codes:
2491  *	0
2492  *	EAGAIN
2493  *		EFAULT
2494  *		ENXIO
2495  *		EIO
2496  */
2497 static int
2498 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2499     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2500     void *cb_arg, vio_desc_direction_t dir)
2501 {
2502 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2503 	int			idx;		/* Index of DRing entry used */
2504 	int			next_idx;
2505 	vio_dring_msg_t		dmsg;
2506 	size_t			msglen;
2507 	int			rv;
2508 
2509 	ASSERT(MUTEX_HELD(&vdcp->lock));
2510 	vdcp->threads_pending++;
2511 loop:
2512 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2513 
2514 	/* Get next available D-Ring entry */
2515 	idx = vdcp->dring_curr_idx;
2516 	local_dep = &(vdcp->local_dring[idx]);
2517 
2518 	if (!local_dep->is_free) {
2519 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
2520 		    vdcp->instance);
2521 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
2522 		if (vdcp->state == VDC_STATE_RUNNING ||
2523 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2524 			goto loop;
2525 		}
2526 		vdcp->threads_pending--;
2527 		return (ECONNRESET);
2528 	}
2529 
2530 	next_idx = idx + 1;
2531 	if (next_idx >= vdcp->dring_len)
2532 		next_idx = 0;
2533 	vdcp->dring_curr_idx = next_idx;
2534 
2535 	ASSERT(local_dep->is_free);
2536 
2537 	local_dep->operation = operation;
2538 	local_dep->addr = addr;
2539 	local_dep->nbytes = nbytes;
2540 	local_dep->slice = slice;
2541 	local_dep->offset = offset;
2542 	local_dep->cb_type = cb_type;
2543 	local_dep->cb_arg = cb_arg;
2544 	local_dep->dir = dir;
2545 
2546 	local_dep->is_free = B_FALSE;
2547 
2548 	rv = vdc_map_to_shared_dring(vdcp, idx);
2549 	if (rv) {
2550 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
2551 		    vdcp->instance);
2552 		/* free the descriptor */
2553 		local_dep->is_free = B_TRUE;
2554 		vdcp->dring_curr_idx = idx;
2555 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
2556 		if (vdcp->state == VDC_STATE_RUNNING ||
2557 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2558 			goto loop;
2559 		}
2560 		vdcp->threads_pending--;
2561 		return (ECONNRESET);
2562 	}
2563 
2564 	/*
2565 	 * Send a msg with the DRing details to vds
2566 	 */
2567 	VIO_INIT_DRING_DATA_TAG(dmsg);
2568 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
2569 	dmsg.dring_ident = vdcp->dring_ident;
2570 	dmsg.start_idx = idx;
2571 	dmsg.end_idx = idx;
2572 	vdcp->seq_num++;
2573 
2574 	DTRACE_IO2(send, vio_dring_msg_t *, &dmsg, vdc_t *, vdcp);
2575 
2576 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
2577 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
2578 
2579 	/*
2580 	 * note we're still holding the lock here to
2581 	 * make sure the message goes out in order !!!...
2582 	 */
2583 	msglen = sizeof (dmsg);
2584 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
2585 	switch (rv) {
2586 	case ECONNRESET:
2587 		/*
2588 		 * vdc_send initiates the reset on failure.
2589 		 * Since the transaction has already been put
2590 		 * on the local dring, it will automatically get
2591 		 * retried when the channel is reset. Given that,
2592 		 * it is ok to just return success even though the
2593 		 * send failed.
2594 		 */
2595 		rv = 0;
2596 		break;
2597 
2598 	case 0: /* EOK */
2599 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
2600 		break;
2601 
2602 	default:
2603 		goto cleanup_and_exit;
2604 	}
2605 
2606 	vdcp->threads_pending--;
2607 	return (rv);
2608 
2609 cleanup_and_exit:
2610 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
2611 	return (ENXIO);
2612 }
2613 
2614 /*
2615  * Function:
2616  *	vdc_do_sync_op
2617  *
2618  * Description:
2619  * 	Wrapper around vdc_populate_descriptor that blocks until the
2620  * 	response to the message is available.
2621  *
2622  * Arguments:
2623  *	vdcp	  - the soft state pointer
2624  *	operation - operation we want vds to perform (VD_OP_XXX)
2625  *	addr	  - address of data buf to be read/written.
2626  *	nbytes	  - number of bytes to read/write
2627  *	slice	  - the disk slice this request is for
2628  *	offset	  - relative disk offset
2629  *	cb_type   - type of call - STRATEGY or SYNC
2630  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2631  *			. mode for ioctl(9e)
2632  *			. LP64 diskaddr_t (block I/O)
2633  *	dir	  - direction of operation (READ/WRITE/BOTH)
2634  *
2635  * Return Codes:
2636  *	0
2637  *	EAGAIN
2638  *		EFAULT
2639  *		ENXIO
2640  *		EIO
2641  */
2642 static int
2643 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
2644     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
2645     vio_desc_direction_t dir)
2646 {
2647 	int status;
2648 
2649 	ASSERT(cb_type == CB_SYNC);
2650 
2651 	/*
2652 	 * Grab the lock, if blocked wait until the server
2653 	 * response causes us to wake up again.
2654 	 */
2655 	mutex_enter(&vdcp->lock);
2656 	vdcp->sync_op_cnt++;
2657 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
2658 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
2659 
2660 	if (vdcp->state == VDC_STATE_DETACH) {
2661 		cv_broadcast(&vdcp->sync_blocked_cv);
2662 		vdcp->sync_op_cnt--;
2663 		mutex_exit(&vdcp->lock);
2664 		return (ENXIO);
2665 	}
2666 
2667 	/* now block anyone other thread entering after us */
2668 	vdcp->sync_op_blocked = B_TRUE;
2669 	vdcp->sync_op_pending = B_TRUE;
2670 	mutex_exit(&vdcp->lock);
2671 
2672 	status = vdc_send_request(vdcp, operation, addr,
2673 	    nbytes, slice, offset, cb_type, cb_arg, dir);
2674 
2675 	mutex_enter(&vdcp->lock);
2676 
2677 	if (status != 0) {
2678 		vdcp->sync_op_pending = B_FALSE;
2679 	} else {
2680 		/*
2681 		 * block until our transaction completes.
2682 		 * Also anyone else waiting also gets to go next.
2683 		 */
2684 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
2685 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
2686 
2687 		DMSG(vdcp, 2, ": operation returned %d\n",
2688 		    vdcp->sync_op_status);
2689 		if (vdcp->state == VDC_STATE_DETACH) {
2690 			vdcp->sync_op_pending = B_FALSE;
2691 			status = ENXIO;
2692 		} else {
2693 			status = vdcp->sync_op_status;
2694 		}
2695 	}
2696 
2697 	vdcp->sync_op_status = 0;
2698 	vdcp->sync_op_blocked = B_FALSE;
2699 	vdcp->sync_op_cnt--;
2700 
2701 	/* signal the next waiting thread */
2702 	cv_signal(&vdcp->sync_blocked_cv);
2703 	mutex_exit(&vdcp->lock);
2704 
2705 	return (status);
2706 }
2707 
2708 
2709 /*
2710  * Function:
2711  *	vdc_drain_response()
2712  *
2713  * Description:
2714  * 	When a guest is panicking, the completion of requests needs to be
2715  * 	handled differently because interrupts are disabled and vdc
2716  * 	will not get messages. We have to poll for the messages instead.
2717  *
2718  * Arguments:
2719  *	vdc	- soft state pointer for this instance of the device driver.
2720  *
2721  * Return Code:
2722  *	0	- Success
2723  */
2724 static int
2725 vdc_drain_response(vdc_t *vdc)
2726 {
2727 	int 			rv, idx, retries;
2728 	size_t			msglen;
2729 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
2730 	vio_dring_msg_t		dmsg;
2731 
2732 	mutex_enter(&vdc->lock);
2733 
2734 	retries = 0;
2735 	for (;;) {
2736 		msglen = sizeof (dmsg);
2737 		rv = ldc_read(vdc->ldc_handle, (caddr_t)&dmsg, &msglen);
2738 		if (rv) {
2739 			rv = EINVAL;
2740 			break;
2741 		}
2742 
2743 		/*
2744 		 * if there are no packets wait and check again
2745 		 */
2746 		if ((rv == 0) && (msglen == 0)) {
2747 			if (retries++ > vdc_dump_retries) {
2748 				rv = EAGAIN;
2749 				break;
2750 			}
2751 
2752 			drv_usecwait(vdc_usec_timeout_dump);
2753 			continue;
2754 		}
2755 
2756 		/*
2757 		 * Ignore all messages that are not ACKs/NACKs to
2758 		 * DRing requests.
2759 		 */
2760 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
2761 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
2762 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
2763 			    dmsg.tag.vio_msgtype,
2764 			    dmsg.tag.vio_subtype,
2765 			    dmsg.tag.vio_subtype_env);
2766 			continue;
2767 		}
2768 
2769 		/*
2770 		 * set the appropriate return value for the current request.
2771 		 */
2772 		switch (dmsg.tag.vio_subtype) {
2773 		case VIO_SUBTYPE_ACK:
2774 			rv = 0;
2775 			break;
2776 		case VIO_SUBTYPE_NACK:
2777 			rv = EAGAIN;
2778 			break;
2779 		default:
2780 			continue;
2781 		}
2782 
2783 		idx = dmsg.start_idx;
2784 		if (idx >= vdc->dring_len) {
2785 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
2786 			    vdc->instance, idx);
2787 			continue;
2788 		}
2789 		ldep = &vdc->local_dring[idx];
2790 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
2791 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
2792 			    vdc->instance, idx, ldep->dep->hdr.dstate);
2793 			continue;
2794 		}
2795 
2796 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
2797 		    vdc->instance, idx, ldep->dep->hdr.dstate);
2798 		rv = vdc_depopulate_descriptor(vdc, idx);
2799 		if (rv) {
2800 			DMSG(vdc, 0,
2801 			    "[%d] Entry @ %d - depopulate failed ..\n",
2802 			    vdc->instance, idx);
2803 		}
2804 
2805 		/* if this is the last descriptor - break out of loop */
2806 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx)
2807 			break;
2808 	}
2809 
2810 	mutex_exit(&vdc->lock);
2811 	DMSG(vdc, 0, "End idx=%d\n", idx);
2812 
2813 	return (rv);
2814 }
2815 
2816 
2817 /*
2818  * Function:
2819  *	vdc_depopulate_descriptor()
2820  *
2821  * Description:
2822  *
2823  * Arguments:
2824  *	vdc	- soft state pointer for this instance of the device driver.
2825  *	idx	- Index of the Descriptor Ring entry being modified
2826  *
2827  * Return Code:
2828  *	0	- Success
2829  */
2830 static int
2831 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
2832 {
2833 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
2834 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
2835 	int		status = ENXIO;
2836 	int		rv = 0;
2837 
2838 	ASSERT(vdc != NULL);
2839 	ASSERT(idx < vdc->dring_len);
2840 	ldep = &vdc->local_dring[idx];
2841 	ASSERT(ldep != NULL);
2842 	ASSERT(MUTEX_HELD(&vdc->lock));
2843 
2844 	DMSG(vdc, 2, ": idx = %d\n", idx);
2845 	dep = ldep->dep;
2846 	ASSERT(dep != NULL);
2847 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
2848 	    (dep->payload.status == ECANCELED));
2849 
2850 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
2851 
2852 	ldep->is_free = B_TRUE;
2853 	status = dep->payload.status;
2854 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
2855 
2856 	/*
2857 	 * If no buffers were used to transfer information to the server when
2858 	 * populating the descriptor then no memory handles need to be unbound
2859 	 * and we can return now.
2860 	 */
2861 	if (ldep->nbytes == 0) {
2862 		cv_signal(&vdc->dring_free_cv);
2863 		return (status);
2864 	}
2865 
2866 	/*
2867 	 * If the upper layer passed in a misaligned address we copied the
2868 	 * data into an aligned buffer before sending it to LDC - we now
2869 	 * copy it back to the original buffer.
2870 	 */
2871 	if (ldep->align_addr) {
2872 		ASSERT(ldep->addr != NULL);
2873 
2874 		if (dep->payload.nbytes > 0)
2875 			bcopy(ldep->align_addr, ldep->addr,
2876 			    dep->payload.nbytes);
2877 		kmem_free(ldep->align_addr,
2878 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
2879 		ldep->align_addr = NULL;
2880 	}
2881 
2882 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
2883 	if (rv != 0) {
2884 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
2885 		    vdc->instance, ldep->desc_mhdl, idx, rv);
2886 		/*
2887 		 * The error returned by the vDisk server is more informative
2888 		 * and thus has a higher priority but if it isn't set we ensure
2889 		 * that this function returns an error.
2890 		 */
2891 		if (status == 0)
2892 			status = EINVAL;
2893 	}
2894 
2895 	cv_signal(&vdc->membind_cv);
2896 	cv_signal(&vdc->dring_free_cv);
2897 
2898 	return (status);
2899 }
2900 
2901 /*
2902  * Function:
2903  *	vdc_populate_mem_hdl()
2904  *
2905  * Description:
2906  *
2907  * Arguments:
2908  *	vdc	- soft state pointer for this instance of the device driver.
2909  *	idx	- Index of the Descriptor Ring entry being modified
2910  *	addr	- virtual address being mapped in
2911  *	nybtes	- number of bytes in 'addr'
2912  *	operation - the vDisk operation being performed (VD_OP_xxx)
2913  *
2914  * Return Code:
2915  *	0	- Success
2916  */
2917 static int
2918 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
2919 {
2920 	vd_dring_entry_t	*dep = NULL;
2921 	ldc_mem_handle_t	mhdl;
2922 	caddr_t			vaddr;
2923 	size_t			nbytes;
2924 	uint8_t			perm = LDC_MEM_RW;
2925 	uint8_t			maptype;
2926 	int			rv = 0;
2927 	int			i;
2928 
2929 	ASSERT(vdcp != NULL);
2930 
2931 	dep = ldep->dep;
2932 	mhdl = ldep->desc_mhdl;
2933 
2934 	switch (ldep->dir) {
2935 	case VIO_read_dir:
2936 		perm = LDC_MEM_W;
2937 		break;
2938 
2939 	case VIO_write_dir:
2940 		perm = LDC_MEM_R;
2941 		break;
2942 
2943 	case VIO_both_dir:
2944 		perm = LDC_MEM_RW;
2945 		break;
2946 
2947 	default:
2948 		ASSERT(0);	/* catch bad programming in vdc */
2949 	}
2950 
2951 	/*
2952 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
2953 	 * to copy the contents of any misaligned buffers to a newly allocated
2954 	 * buffer and bind it instead (and copy the the contents back to the
2955 	 * original buffer passed in when depopulating the descriptor)
2956 	 */
2957 	vaddr = ldep->addr;
2958 	nbytes = ldep->nbytes;
2959 	if (((uint64_t)vaddr & 0x7) != 0) {
2960 		ASSERT(ldep->align_addr == NULL);
2961 		ldep->align_addr =
2962 		    kmem_alloc(sizeof (caddr_t) *
2963 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
2964 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
2965 		    "(buf=%p nb=%ld op=%d)\n",
2966 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
2967 		    nbytes, ldep->operation);
2968 		if (perm != LDC_MEM_W)
2969 			bcopy(vaddr, ldep->align_addr, nbytes);
2970 		vaddr = ldep->align_addr;
2971 	}
2972 
2973 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
2974 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
2975 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
2976 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
2977 	    vdcp->instance, dep->payload.ncookies);
2978 	if (rv != 0) {
2979 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
2980 		    "(mhdl=%p, buf=%p, err=%d)\n",
2981 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
2982 		if (ldep->align_addr) {
2983 			kmem_free(ldep->align_addr,
2984 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
2985 			ldep->align_addr = NULL;
2986 		}
2987 		return (EAGAIN);
2988 	}
2989 
2990 	/*
2991 	 * Get the other cookies (if any).
2992 	 */
2993 	for (i = 1; i < dep->payload.ncookies; i++) {
2994 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
2995 		if (rv != 0) {
2996 			(void) ldc_mem_unbind_handle(mhdl);
2997 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
2998 			    "(mhdl=%lx cnum=%d), err=%d",
2999 			    vdcp->instance, mhdl, i, rv);
3000 			if (ldep->align_addr) {
3001 				kmem_free(ldep->align_addr,
3002 				    sizeof (caddr_t) * ldep->nbytes);
3003 				ldep->align_addr = NULL;
3004 			}
3005 			return (EAGAIN);
3006 		}
3007 	}
3008 
3009 	return (rv);
3010 }
3011 
3012 /*
3013  * Interrupt handlers for messages from LDC
3014  */
3015 
3016 /*
3017  * Function:
3018  *	vdc_handle_cb()
3019  *
3020  * Description:
3021  *
3022  * Arguments:
3023  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3024  *	arg	- soft state pointer for this instance of the device driver.
3025  *
3026  * Return Code:
3027  *	0	- Success
3028  */
3029 static uint_t
3030 vdc_handle_cb(uint64_t event, caddr_t arg)
3031 {
3032 	ldc_status_t	ldc_state;
3033 	int		rv = 0;
3034 
3035 	vdc_t	*vdc = (vdc_t *)(void *)arg;
3036 
3037 	ASSERT(vdc != NULL);
3038 
3039 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3040 
3041 	/*
3042 	 * Depending on the type of event that triggered this callback,
3043 	 * we modify the handshake state or read the data.
3044 	 *
3045 	 * NOTE: not done as a switch() as event could be triggered by
3046 	 * a state change and a read request. Also the ordering	of the
3047 	 * check for the event types is deliberate.
3048 	 */
3049 	if (event & LDC_EVT_UP) {
3050 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3051 
3052 		mutex_enter(&vdc->lock);
3053 
3054 		/* get LDC state */
3055 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
3056 		if (rv != 0) {
3057 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3058 			    vdc->instance, rv);
3059 			return (LDC_SUCCESS);
3060 		}
3061 		if (vdc->ldc_state != LDC_UP && ldc_state == LDC_UP) {
3062 			/*
3063 			 * Reset the transaction sequence numbers when
3064 			 * LDC comes up. We then kick off the handshake
3065 			 * negotiation with the vDisk server.
3066 			 */
3067 			vdc->seq_num = 1;
3068 			vdc->seq_num_reply = 0;
3069 			vdc->ldc_state = ldc_state;
3070 			cv_signal(&vdc->initwait_cv);
3071 		}
3072 
3073 		mutex_exit(&vdc->lock);
3074 	}
3075 
3076 	if (event & LDC_EVT_READ) {
3077 		DMSG(vdc, 0, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3078 		mutex_enter(&vdc->read_lock);
3079 		cv_signal(&vdc->read_cv);
3080 		vdc->read_state = VDC_READ_PENDING;
3081 		mutex_exit(&vdc->read_lock);
3082 
3083 		/* that's all we have to do - no need to handle DOWN/RESET */
3084 		return (LDC_SUCCESS);
3085 	}
3086 
3087 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3088 
3089 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3090 
3091 		mutex_enter(&vdc->lock);
3092 		/*
3093 		 * Need to wake up any readers so they will
3094 		 * detect that a reset has occurred.
3095 		 */
3096 		mutex_enter(&vdc->read_lock);
3097 		if ((vdc->read_state == VDC_READ_WAITING) ||
3098 		    (vdc->read_state == VDC_READ_RESET))
3099 			cv_signal(&vdc->read_cv);
3100 		vdc->read_state = VDC_READ_RESET;
3101 		mutex_exit(&vdc->read_lock);
3102 
3103 		/* wake up any threads waiting for connection to come up */
3104 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3105 			vdc->state = VDC_STATE_RESETTING;
3106 			cv_signal(&vdc->initwait_cv);
3107 		}
3108 
3109 		mutex_exit(&vdc->lock);
3110 	}
3111 
3112 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3113 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3114 		    vdc->instance, event);
3115 
3116 	return (LDC_SUCCESS);
3117 }
3118 
3119 /*
3120  * Function:
3121  *	vdc_wait_for_response()
3122  *
3123  * Description:
3124  *	Block waiting for a response from the server. If there is
3125  *	no data the thread block on the read_cv that is signalled
3126  *	by the callback when an EVT_READ occurs.
3127  *
3128  * Arguments:
3129  *	vdcp	- soft state pointer for this instance of the device driver.
3130  *
3131  * Return Code:
3132  *	0	- Success
3133  */
3134 static int
3135 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3136 {
3137 	size_t		nbytes = sizeof (*msgp);
3138 	int		status;
3139 
3140 	ASSERT(vdcp != NULL);
3141 
3142 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3143 
3144 	status = vdc_recv(vdcp, msgp, &nbytes);
3145 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3146 	    status, (int)nbytes);
3147 	if (status) {
3148 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3149 		    vdcp->instance, status);
3150 		return (status);
3151 	}
3152 
3153 	if (nbytes < sizeof (vio_msg_tag_t)) {
3154 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3155 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3156 		return (ENOMSG);
3157 	}
3158 
3159 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3160 	    msgp->tag.vio_msgtype,
3161 	    msgp->tag.vio_subtype,
3162 	    msgp->tag.vio_subtype_env);
3163 
3164 	/*
3165 	 * Verify the Session ID of the message
3166 	 *
3167 	 * Every message after the Version has been negotiated should
3168 	 * have the correct session ID set.
3169 	 */
3170 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3171 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3172 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3173 		    "expected 0x%lx [seq num %lx @ %d]",
3174 		    vdcp->instance, msgp->tag.vio_sid,
3175 		    vdcp->session_id,
3176 		    ((vio_dring_msg_t *)msgp)->seq_num,
3177 		    ((vio_dring_msg_t *)msgp)->start_idx);
3178 		return (ENOMSG);
3179 	}
3180 	return (0);
3181 }
3182 
3183 
3184 /*
3185  * Function:
3186  *	vdc_resubmit_backup_dring()
3187  *
3188  * Description:
3189  *	Resubmit each descriptor in the backed up dring to
3190  * 	vDisk server. The Dring was backed up during connection
3191  *	reset.
3192  *
3193  * Arguments:
3194  *	vdcp	- soft state pointer for this instance of the device driver.
3195  *
3196  * Return Code:
3197  *	0	- Success
3198  */
3199 static int
3200 vdc_resubmit_backup_dring(vdc_t *vdcp)
3201 {
3202 	int		count;
3203 	int		b_idx;
3204 	int		rv;
3205 	int		dring_size;
3206 	int		status;
3207 	vio_msg_t	vio_msg;
3208 	vdc_local_desc_t	*curr_ldep;
3209 
3210 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3211 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3212 
3213 	if (vdcp->local_dring_backup == NULL) {
3214 		/* the pending requests have already been processed */
3215 		return (0);
3216 	}
3217 
3218 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3219 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3220 
3221 	/*
3222 	 * Walk the backup copy of the local descriptor ring and
3223 	 * resubmit all the outstanding transactions.
3224 	 */
3225 	b_idx = vdcp->local_dring_backup_tail;
3226 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3227 
3228 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3229 
3230 		/* only resubmit outstanding transactions */
3231 		if (!curr_ldep->is_free) {
3232 
3233 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3234 			mutex_enter(&vdcp->lock);
3235 			rv = vdc_populate_descriptor(vdcp, curr_ldep->operation,
3236 			    curr_ldep->addr, curr_ldep->nbytes,
3237 			    curr_ldep->slice, curr_ldep->offset,
3238 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3239 			    curr_ldep->dir);
3240 			mutex_exit(&vdcp->lock);
3241 			if (rv) {
3242 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3243 				    vdcp->instance, b_idx);
3244 				return (rv);
3245 			}
3246 
3247 			/* Wait for the response message. */
3248 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3249 			    b_idx);
3250 			status = vdc_wait_for_response(vdcp, &vio_msg);
3251 			if (status) {
3252 				DMSG(vdcp, 1, "[%d] wait_for_response "
3253 				    "returned err=%d\n", vdcp->instance,
3254 				    status);
3255 				return (status);
3256 			}
3257 
3258 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3259 			status = vdc_process_data_msg(vdcp, &vio_msg);
3260 			if (status) {
3261 				DMSG(vdcp, 1, "[%d] process_data_msg "
3262 				    "returned err=%d\n", vdcp->instance,
3263 				    status);
3264 				return (status);
3265 			}
3266 		}
3267 
3268 		/* get the next element to submit */
3269 		if (++b_idx >= vdcp->local_dring_backup_len)
3270 			b_idx = 0;
3271 	}
3272 
3273 	/* all done - now clear up pending dring copy */
3274 	dring_size = vdcp->local_dring_backup_len *
3275 	    sizeof (vdcp->local_dring_backup[0]);
3276 
3277 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3278 
3279 	vdcp->local_dring_backup = NULL;
3280 
3281 	return (0);
3282 }
3283 
3284 /*
3285  * Function:
3286  *	vdc_cancel_backup_dring
3287  *
3288  * Description:
3289  *	Cancel each descriptor in the backed up dring to vDisk server.
3290  *	The Dring was backed up during connection reset.
3291  *
3292  * Arguments:
3293  *	vdcp	- soft state pointer for this instance of the device driver.
3294  *
3295  * Return Code:
3296  *	None
3297  */
3298 void
3299 vdc_cancel_backup_ring(vdc_t *vdcp)
3300 {
3301 	vdc_local_desc_t *ldep;
3302 	struct buf 	*bufp;
3303 	int		count;
3304 	int		b_idx;
3305 	int		dring_size;
3306 
3307 	ASSERT(MUTEX_HELD(&vdcp->lock));
3308 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3309 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3310 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3311 	    vdcp->state == VDC_STATE_RESETTING);
3312 
3313 	if (vdcp->local_dring_backup == NULL) {
3314 		/* the pending requests have already been processed */
3315 		return;
3316 	}
3317 
3318 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3319 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3320 
3321 	/*
3322 	 * Walk the backup copy of the local descriptor ring and
3323 	 * cancel all the outstanding transactions.
3324 	 */
3325 	b_idx = vdcp->local_dring_backup_tail;
3326 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3327 
3328 		ldep = &(vdcp->local_dring_backup[b_idx]);
3329 
3330 		/* only cancel outstanding transactions */
3331 		if (!ldep->is_free) {
3332 
3333 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3334 
3335 			/*
3336 			 * All requests have already been cleared from the
3337 			 * local descriptor ring and the LDC channel has been
3338 			 * reset so we will never get any reply for these
3339 			 * requests. Now we just have to notify threads waiting
3340 			 * for replies that the request has failed.
3341 			 */
3342 			switch (ldep->cb_type) {
3343 			case CB_SYNC:
3344 				ASSERT(vdcp->sync_op_pending);
3345 				vdcp->sync_op_status = EIO;
3346 				vdcp->sync_op_pending = B_FALSE;
3347 				cv_signal(&vdcp->sync_pending_cv);
3348 				break;
3349 
3350 			case CB_STRATEGY:
3351 				bufp = ldep->cb_arg;
3352 				ASSERT(bufp != NULL);
3353 				bufp->b_resid = bufp->b_bcount;
3354 				bioerror(bufp, EIO);
3355 				biodone(bufp);
3356 				break;
3357 
3358 			default:
3359 				ASSERT(0);
3360 			}
3361 
3362 		}
3363 
3364 		/* get the next element to cancel */
3365 		if (++b_idx >= vdcp->local_dring_backup_len)
3366 			b_idx = 0;
3367 	}
3368 
3369 	/* all done - now clear up pending dring copy */
3370 	dring_size = vdcp->local_dring_backup_len *
3371 	    sizeof (vdcp->local_dring_backup[0]);
3372 
3373 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3374 
3375 	vdcp->local_dring_backup = NULL;
3376 
3377 	DTRACE_IO2(processed, int, count, vdc_t *, vdcp);
3378 }
3379 
3380 /*
3381  * Function:
3382  *	vdc_connection_timeout
3383  *
3384  * Description:
3385  *	This function is invoked if the timeout set to establish the connection
3386  *	with vds expires. This will happen if we spend too much time in the
3387  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
3388  *	cancel any pending request and mark them as failed.
3389  *
3390  *	If the timeout does not expire, it will be cancelled when we reach the
3391  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
3392  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
3393  *	VDC_STATE_RESETTING state in which case we do nothing because the
3394  *	timeout is being cancelled.
3395  *
3396  * Arguments:
3397  *	arg	- argument of the timeout function actually a soft state
3398  *		  pointer for the instance of the device driver.
3399  *
3400  * Return Code:
3401  *	None
3402  */
3403 void
3404 vdc_connection_timeout(void *arg)
3405 {
3406 	vdc_t 		*vdcp = (vdc_t *)arg;
3407 
3408 	mutex_enter(&vdcp->lock);
3409 
3410 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
3411 	    vdcp->state == VDC_STATE_DETACH) {
3412 		/*
3413 		 * The connection has just been re-established or
3414 		 * we are detaching.
3415 		 */
3416 		vdcp->ctimeout_reached = B_FALSE;
3417 		mutex_exit(&vdcp->lock);
3418 		return;
3419 	}
3420 
3421 	vdcp->ctimeout_reached = B_TRUE;
3422 
3423 	/* notify requests waiting for sending */
3424 	cv_broadcast(&vdcp->running_cv);
3425 
3426 	/* cancel requests waiting for a result */
3427 	vdc_cancel_backup_ring(vdcp);
3428 
3429 	mutex_exit(&vdcp->lock);
3430 
3431 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
3432 	    vdcp->instance);
3433 }
3434 
3435 /*
3436  * Function:
3437  *	vdc_backup_local_dring()
3438  *
3439  * Description:
3440  *	Backup the current dring in the event of a reset. The Dring
3441  *	transactions will be resubmitted to the server when the
3442  *	connection is restored.
3443  *
3444  * Arguments:
3445  *	vdcp	- soft state pointer for this instance of the device driver.
3446  *
3447  * Return Code:
3448  *	NONE
3449  */
3450 static void
3451 vdc_backup_local_dring(vdc_t *vdcp)
3452 {
3453 	int dring_size;
3454 
3455 	ASSERT(MUTEX_HELD(&vdcp->lock));
3456 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
3457 
3458 	/*
3459 	 * If the backup dring is stil around, it means
3460 	 * that the last restore did not complete. However,
3461 	 * since we never got back into the running state,
3462 	 * the backup copy we have is still valid.
3463 	 */
3464 	if (vdcp->local_dring_backup != NULL) {
3465 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
3466 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
3467 		    vdcp->local_dring_backup_tail);
3468 		return;
3469 	}
3470 
3471 	/*
3472 	 * The backup dring can be NULL and the local dring may not be
3473 	 * initialized. This can happen if we had a reset while establishing
3474 	 * a new connection but after the connection has timed out. In that
3475 	 * case the backup dring is NULL because the requests have been
3476 	 * cancelled and the request occured before the local dring is
3477 	 * initialized.
3478 	 */
3479 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
3480 		return;
3481 
3482 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
3483 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
3484 
3485 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
3486 
3487 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
3488 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
3489 
3490 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
3491 	vdcp->local_dring_backup_len = vdcp->dring_len;
3492 }
3493 
3494 /* -------------------------------------------------------------------------- */
3495 
3496 /*
3497  * The following functions process the incoming messages from vds
3498  */
3499 
3500 /*
3501  * Function:
3502  *      vdc_process_msg_thread()
3503  *
3504  * Description:
3505  *
3506  *	Main VDC message processing thread. Each vDisk instance
3507  * 	consists of a copy of this thread. This thread triggers
3508  * 	all the handshakes and data exchange with the server. It
3509  * 	also handles all channel resets
3510  *
3511  * Arguments:
3512  *      vdc     - soft state pointer for this instance of the device driver.
3513  *
3514  * Return Code:
3515  *      None
3516  */
3517 static void
3518 vdc_process_msg_thread(vdc_t *vdcp)
3519 {
3520 	int	status;
3521 	int	ctimeout;
3522 	timeout_id_t tmid = 0;
3523 
3524 	mutex_enter(&vdcp->lock);
3525 
3526 	for (;;) {
3527 
3528 #define	Q(_s)	(vdcp->state == _s) ? #_s :
3529 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
3530 		    Q(VDC_STATE_INIT)
3531 		    Q(VDC_STATE_INIT_WAITING)
3532 		    Q(VDC_STATE_NEGOTIATE)
3533 		    Q(VDC_STATE_HANDLE_PENDING)
3534 		    Q(VDC_STATE_RUNNING)
3535 		    Q(VDC_STATE_RESETTING)
3536 		    Q(VDC_STATE_DETACH)
3537 		    "UNKNOWN");
3538 
3539 		switch (vdcp->state) {
3540 		case VDC_STATE_INIT:
3541 
3542 			/*
3543 			 * If requested, start a timeout to check if the
3544 			 * connection with vds is established in the
3545 			 * specified delay. If the timeout expires, we
3546 			 * will cancel any pending request.
3547 			 *
3548 			 * If some reset have occurred while establishing
3549 			 * the connection, we already have a timeout armed
3550 			 * and in that case we don't need to arm a new one.
3551 			 */
3552 			ctimeout = (vdc_timeout != 0)?
3553 			    vdc_timeout : vdcp->ctimeout;
3554 
3555 			if (ctimeout != 0 && tmid == 0) {
3556 				tmid = timeout(vdc_connection_timeout, vdcp,
3557 				    ctimeout * drv_usectohz(1000000));
3558 			}
3559 
3560 			/* Check if have re-initializing repeatedly */
3561 			if (vdcp->hshake_cnt++ > vdc_hshake_retries &&
3562 			    vdcp->lifecycle != VDC_LC_ONLINE) {
3563 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
3564 				    vdcp->instance);
3565 				vdcp->state = VDC_STATE_DETACH;
3566 				break;
3567 			}
3568 
3569 			/* Bring up connection with vds via LDC */
3570 			status = vdc_start_ldc_connection(vdcp);
3571 			if (status == EINVAL) {
3572 				DMSG(vdcp, 0, "[%d] Could not start LDC",
3573 				    vdcp->instance);
3574 				vdcp->state = VDC_STATE_DETACH;
3575 			} else {
3576 				vdcp->state = VDC_STATE_INIT_WAITING;
3577 			}
3578 			break;
3579 
3580 		case VDC_STATE_INIT_WAITING:
3581 
3582 			/*
3583 			 * Let the callback event move us on
3584 			 * when channel is open to server
3585 			 */
3586 			while (vdcp->ldc_state != LDC_UP) {
3587 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
3588 				if (vdcp->state != VDC_STATE_INIT_WAITING) {
3589 					DMSG(vdcp, 0,
3590 				"state moved to %d out from under us...\n",
3591 					    vdcp->state);
3592 
3593 					break;
3594 				}
3595 			}
3596 			if (vdcp->state == VDC_STATE_INIT_WAITING &&
3597 			    vdcp->ldc_state == LDC_UP) {
3598 				vdcp->state = VDC_STATE_NEGOTIATE;
3599 			}
3600 			break;
3601 
3602 		case VDC_STATE_NEGOTIATE:
3603 			switch (status = vdc_ver_negotiation(vdcp)) {
3604 			case 0:
3605 				break;
3606 			default:
3607 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
3608 				    status);
3609 				goto reset;
3610 			}
3611 
3612 			switch (status = vdc_attr_negotiation(vdcp)) {
3613 			case 0:
3614 				break;
3615 			default:
3616 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
3617 				    status);
3618 				goto reset;
3619 			}
3620 
3621 			switch (status = vdc_dring_negotiation(vdcp)) {
3622 			case 0:
3623 				break;
3624 			default:
3625 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
3626 				    status);
3627 				goto reset;
3628 			}
3629 
3630 			switch (status = vdc_rdx_exchange(vdcp)) {
3631 			case 0:
3632 				vdcp->state = VDC_STATE_HANDLE_PENDING;
3633 				goto done;
3634 			default:
3635 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
3636 				    status);
3637 				goto reset;
3638 			}
3639 reset:
3640 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
3641 			    status);
3642 			vdcp->state = VDC_STATE_RESETTING;
3643 			vdcp->self_reset = B_TRUE;
3644 done:
3645 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
3646 			    vdcp->state);
3647 			break;
3648 
3649 		case VDC_STATE_HANDLE_PENDING:
3650 
3651 			if (vdcp->ctimeout_reached) {
3652 				/*
3653 				 * The connection timeout had been reached so
3654 				 * pending requests have been cancelled. Now
3655 				 * that the connection is back we can reset
3656 				 * the timeout.
3657 				 */
3658 				ASSERT(vdcp->local_dring_backup == NULL);
3659 				ASSERT(tmid != 0);
3660 				tmid = 0;
3661 				vdcp->ctimeout_reached = B_FALSE;
3662 				vdcp->state = VDC_STATE_RUNNING;
3663 				DMSG(vdcp, 0, "[%d] connection to service "
3664 				    "domain is up", vdcp->instance);
3665 				break;
3666 			}
3667 
3668 			mutex_exit(&vdcp->lock);
3669 			if (tmid != 0) {
3670 				(void) untimeout(tmid);
3671 				tmid = 0;
3672 			}
3673 			status = vdc_resubmit_backup_dring(vdcp);
3674 			mutex_enter(&vdcp->lock);
3675 
3676 			if (status)
3677 				vdcp->state = VDC_STATE_RESETTING;
3678 			else
3679 				vdcp->state = VDC_STATE_RUNNING;
3680 
3681 			break;
3682 
3683 		/* enter running state */
3684 		case VDC_STATE_RUNNING:
3685 			/*
3686 			 * Signal anyone waiting for the connection
3687 			 * to come on line.
3688 			 */
3689 			vdcp->hshake_cnt = 0;
3690 			cv_broadcast(&vdcp->running_cv);
3691 			mutex_exit(&vdcp->lock);
3692 
3693 			for (;;) {
3694 				vio_msg_t msg;
3695 				status = vdc_wait_for_response(vdcp, &msg);
3696 				if (status) break;
3697 
3698 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
3699 				    vdcp->instance);
3700 				status = vdc_process_data_msg(vdcp, &msg);
3701 				if (status) {
3702 					DMSG(vdcp, 1, "[%d] process_data_msg "
3703 					    "returned err=%d\n", vdcp->instance,
3704 					    status);
3705 					break;
3706 				}
3707 
3708 			}
3709 
3710 			mutex_enter(&vdcp->lock);
3711 
3712 			vdcp->state = VDC_STATE_RESETTING;
3713 			vdcp->self_reset = B_TRUE;
3714 			break;
3715 
3716 		case VDC_STATE_RESETTING:
3717 			/*
3718 			 * When we reach this state, we either come from the
3719 			 * VDC_STATE_RUNNING state and we can have pending
3720 			 * request but no timeout is armed; or we come from
3721 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
3722 			 * VDC_HANDLE_PENDING state and there is no pending
3723 			 * request or pending requests have already been copied
3724 			 * into the backup dring. So we can safely keep the
3725 			 * connection timeout armed while we are in this state.
3726 			 */
3727 
3728 			DMSG(vdcp, 0, "Initiating channel reset "
3729 			    "(pending = %d)\n", (int)vdcp->threads_pending);
3730 
3731 			if (vdcp->self_reset) {
3732 				DMSG(vdcp, 0,
3733 				    "[%d] calling stop_ldc_connection.\n",
3734 				    vdcp->instance);
3735 				status = vdc_stop_ldc_connection(vdcp);
3736 				vdcp->self_reset = B_FALSE;
3737 			}
3738 
3739 			/*
3740 			 * Wait for all threads currently waiting
3741 			 * for a free dring entry to use.
3742 			 */
3743 			while (vdcp->threads_pending) {
3744 				cv_broadcast(&vdcp->membind_cv);
3745 				cv_broadcast(&vdcp->dring_free_cv);
3746 				mutex_exit(&vdcp->lock);
3747 				/* give the waiters enough time to wake up */
3748 				delay(vdc_hz_min_ldc_delay);
3749 				mutex_enter(&vdcp->lock);
3750 			}
3751 
3752 			ASSERT(vdcp->threads_pending == 0);
3753 
3754 			/* Sanity check that no thread is receiving */
3755 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
3756 
3757 			vdcp->read_state = VDC_READ_IDLE;
3758 
3759 			vdc_backup_local_dring(vdcp);
3760 
3761 			/* cleanup the old d-ring */
3762 			vdc_destroy_descriptor_ring(vdcp);
3763 
3764 			/* go and start again */
3765 			vdcp->state = VDC_STATE_INIT;
3766 
3767 			break;
3768 
3769 		case VDC_STATE_DETACH:
3770 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
3771 			    vdcp->instance);
3772 
3773 			/* cancel any pending timeout */
3774 			mutex_exit(&vdcp->lock);
3775 			if (tmid != 0) {
3776 				(void) untimeout(tmid);
3777 				tmid = 0;
3778 			}
3779 			mutex_enter(&vdcp->lock);
3780 
3781 			/*
3782 			 * Signal anyone waiting for connection
3783 			 * to come online
3784 			 */
3785 			cv_broadcast(&vdcp->running_cv);
3786 
3787 			while (vdcp->sync_op_pending) {
3788 				cv_signal(&vdcp->sync_pending_cv);
3789 				cv_signal(&vdcp->sync_blocked_cv);
3790 				mutex_exit(&vdcp->lock);
3791 				/* give the waiters enough time to wake up */
3792 				delay(vdc_hz_min_ldc_delay);
3793 				mutex_enter(&vdcp->lock);
3794 			}
3795 
3796 			mutex_exit(&vdcp->lock);
3797 
3798 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
3799 			    vdcp->instance);
3800 			thread_exit();
3801 			break;
3802 		}
3803 	}
3804 }
3805 
3806 
3807 /*
3808  * Function:
3809  *	vdc_process_data_msg()
3810  *
3811  * Description:
3812  *	This function is called by the message processing thread each time
3813  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
3814  *	be an ACK or NACK from vds[1] which vdc handles as follows.
3815  *		ACK	- wake up the waiting thread
3816  *		NACK	- resend any messages necessary
3817  *
3818  *	[1] Although the message format allows it, vds should not send a
3819  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
3820  *	    some bizarre reason it does, vdc will reset the connection.
3821  *
3822  * Arguments:
3823  *	vdc	- soft state pointer for this instance of the device driver.
3824  *	msg	- the LDC message sent by vds
3825  *
3826  * Return Code:
3827  *	0	- Success.
3828  *	> 0	- error value returned by LDC
3829  */
3830 static int
3831 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
3832 {
3833 	int			status = 0;
3834 	vio_dring_msg_t		*dring_msg;
3835 	vdc_local_desc_t	*ldep = NULL;
3836 	int			start, end;
3837 	int			idx;
3838 
3839 	dring_msg = (vio_dring_msg_t *)msg;
3840 
3841 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
3842 	ASSERT(vdcp != NULL);
3843 
3844 	mutex_enter(&vdcp->lock);
3845 
3846 	/*
3847 	 * Check to see if the message has bogus data
3848 	 */
3849 	idx = start = dring_msg->start_idx;
3850 	end = dring_msg->end_idx;
3851 	if ((start >= vdcp->dring_len) ||
3852 	    (end >= vdcp->dring_len) || (end < -1)) {
3853 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
3854 		    vdcp->instance, start, end);
3855 		mutex_exit(&vdcp->lock);
3856 		return (EINVAL);
3857 	}
3858 
3859 	/*
3860 	 * Verify that the sequence number is what vdc expects.
3861 	 */
3862 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
3863 	case VDC_SEQ_NUM_TODO:
3864 		break;	/* keep processing this message */
3865 	case VDC_SEQ_NUM_SKIP:
3866 		mutex_exit(&vdcp->lock);
3867 		return (0);
3868 	case VDC_SEQ_NUM_INVALID:
3869 		mutex_exit(&vdcp->lock);
3870 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
3871 		return (ENXIO);
3872 	}
3873 
3874 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
3875 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
3876 		VDC_DUMP_DRING_MSG(dring_msg);
3877 		mutex_exit(&vdcp->lock);
3878 		return (EIO);
3879 
3880 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
3881 		mutex_exit(&vdcp->lock);
3882 		return (EPROTO);
3883 	}
3884 
3885 	DTRACE_IO2(recv, vio_dring_msg_t, dring_msg, vdc_t *, vdcp);
3886 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
3887 	ASSERT(start == end);
3888 
3889 	ldep = &vdcp->local_dring[idx];
3890 
3891 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
3892 	    ldep->dep->hdr.dstate, ldep->cb_type);
3893 
3894 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
3895 		struct buf *bufp;
3896 
3897 		switch (ldep->cb_type) {
3898 		case CB_SYNC:
3899 			ASSERT(vdcp->sync_op_pending);
3900 
3901 			status = vdc_depopulate_descriptor(vdcp, idx);
3902 			vdcp->sync_op_status = status;
3903 			vdcp->sync_op_pending = B_FALSE;
3904 			cv_signal(&vdcp->sync_pending_cv);
3905 			break;
3906 
3907 		case CB_STRATEGY:
3908 			bufp = ldep->cb_arg;
3909 			ASSERT(bufp != NULL);
3910 			bufp->b_resid =
3911 			    bufp->b_bcount - ldep->dep->payload.nbytes;
3912 			status = ldep->dep->payload.status; /* Future:ntoh */
3913 			if (status != 0) {
3914 				DMSG(vdcp, 1, "strategy status=%d\n", status);
3915 				bioerror(bufp, status);
3916 			}
3917 			status = vdc_depopulate_descriptor(vdcp, idx);
3918 			biodone(bufp);
3919 
3920 			DMSG(vdcp, 1,
3921 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
3922 			    bufp->b_bcount, ldep->dep->payload.nbytes);
3923 			break;
3924 
3925 		default:
3926 			ASSERT(0);
3927 		}
3928 	}
3929 
3930 	/* let the arrival signal propogate */
3931 	mutex_exit(&vdcp->lock);
3932 
3933 	/* probe gives the count of how many entries were processed */
3934 	DTRACE_IO2(processed, int, 1, vdc_t *, vdcp);
3935 
3936 	return (0);
3937 }
3938 
3939 /*
3940  * Function:
3941  *	vdc_process_err_msg()
3942  *
3943  * NOTE: No error messages are used as part of the vDisk protocol
3944  */
3945 static int
3946 vdc_process_err_msg(vdc_t *vdc, vio_msg_t msg)
3947 {
3948 	_NOTE(ARGUNUSED(vdc))
3949 	_NOTE(ARGUNUSED(msg))
3950 
3951 	ASSERT(msg.tag.vio_msgtype == VIO_TYPE_ERR);
3952 	DMSG(vdc, 1, "[%d] Got an ERR msg", vdc->instance);
3953 
3954 	return (ENOTSUP);
3955 }
3956 
3957 /*
3958  * Function:
3959  *	vdc_handle_ver_msg()
3960  *
3961  * Description:
3962  *
3963  * Arguments:
3964  *	vdc	- soft state pointer for this instance of the device driver.
3965  *	ver_msg	- LDC message sent by vDisk server
3966  *
3967  * Return Code:
3968  *	0	- Success
3969  */
3970 static int
3971 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
3972 {
3973 	int status = 0;
3974 
3975 	ASSERT(vdc != NULL);
3976 	ASSERT(mutex_owned(&vdc->lock));
3977 
3978 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
3979 		return (EPROTO);
3980 	}
3981 
3982 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
3983 		return (EINVAL);
3984 	}
3985 
3986 	switch (ver_msg->tag.vio_subtype) {
3987 	case VIO_SUBTYPE_ACK:
3988 		/*
3989 		 * We check to see if the version returned is indeed supported
3990 		 * (The server may have also adjusted the minor number downwards
3991 		 * and if so 'ver_msg' will contain the actual version agreed)
3992 		 */
3993 		if (vdc_is_supported_version(ver_msg)) {
3994 			vdc->ver.major = ver_msg->ver_major;
3995 			vdc->ver.minor = ver_msg->ver_minor;
3996 			ASSERT(vdc->ver.major > 0);
3997 		} else {
3998 			status = EPROTO;
3999 		}
4000 		break;
4001 
4002 	case VIO_SUBTYPE_NACK:
4003 		/*
4004 		 * call vdc_is_supported_version() which will return the next
4005 		 * supported version (if any) in 'ver_msg'
4006 		 */
4007 		(void) vdc_is_supported_version(ver_msg);
4008 		if (ver_msg->ver_major > 0) {
4009 			size_t len = sizeof (*ver_msg);
4010 
4011 			ASSERT(vdc->ver.major > 0);
4012 
4013 			/* reset the necessary fields and resend */
4014 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4015 			ver_msg->dev_class = VDEV_DISK;
4016 
4017 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4018 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4019 			    vdc->instance, status);
4020 			if (len != sizeof (*ver_msg))
4021 				status = EBADMSG;
4022 		} else {
4023 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4024 			    vdc->instance);
4025 			status = ENOTSUP;
4026 		}
4027 
4028 		break;
4029 	case VIO_SUBTYPE_INFO:
4030 		/*
4031 		 * Handle the case where vds starts handshake
4032 		 * (for now only vdc is the instigator)
4033 		 */
4034 		status = ENOTSUP;
4035 		break;
4036 
4037 	default:
4038 		status = EINVAL;
4039 		break;
4040 	}
4041 
4042 	return (status);
4043 }
4044 
4045 /*
4046  * Function:
4047  *	vdc_handle_attr_msg()
4048  *
4049  * Description:
4050  *
4051  * Arguments:
4052  *	vdc	- soft state pointer for this instance of the device driver.
4053  *	attr_msg	- LDC message sent by vDisk server
4054  *
4055  * Return Code:
4056  *	0	- Success
4057  */
4058 static int
4059 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4060 {
4061 	int status = 0;
4062 
4063 	ASSERT(vdc != NULL);
4064 	ASSERT(mutex_owned(&vdc->lock));
4065 
4066 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4067 		return (EPROTO);
4068 	}
4069 
4070 	switch (attr_msg->tag.vio_subtype) {
4071 	case VIO_SUBTYPE_ACK:
4072 		/*
4073 		 * We now verify the attributes sent by vds.
4074 		 */
4075 		vdc->vdisk_size = attr_msg->vdisk_size;
4076 		vdc->vdisk_type = attr_msg->vdisk_type;
4077 
4078 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4079 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4080 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4081 		    vdc->instance, vdc->block_size,
4082 		    attr_msg->vdisk_block_size);
4083 
4084 		/*
4085 		 * We don't know at compile time what the vDisk server will
4086 		 * think are good values but we apply an large (arbitrary)
4087 		 * upper bound to prevent memory exhaustion in vdc if it was
4088 		 * allocating a DRing based of huge values sent by the server.
4089 		 * We probably will never exceed this except if the message
4090 		 * was garbage.
4091 		 */
4092 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4093 		    (PAGESIZE * DEV_BSIZE)) {
4094 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4095 			vdc->block_size = attr_msg->vdisk_block_size;
4096 		} else {
4097 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4098 			    " using max supported by vdc", vdc->instance);
4099 		}
4100 
4101 		if ((attr_msg->xfer_mode != VIO_DRING_MODE) ||
4102 		    (attr_msg->vdisk_size > INT64_MAX) ||
4103 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4104 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4105 			    vdc->instance);
4106 			status = EINVAL;
4107 			break;
4108 		}
4109 
4110 		break;
4111 
4112 	case VIO_SUBTYPE_NACK:
4113 		/*
4114 		 * vds could not handle the attributes we sent so we
4115 		 * stop negotiating.
4116 		 */
4117 		status = EPROTO;
4118 		break;
4119 
4120 	case VIO_SUBTYPE_INFO:
4121 		/*
4122 		 * Handle the case where vds starts the handshake
4123 		 * (for now; vdc is the only supported instigatior)
4124 		 */
4125 		status = ENOTSUP;
4126 		break;
4127 
4128 	default:
4129 		status = ENOTSUP;
4130 		break;
4131 	}
4132 
4133 	return (status);
4134 }
4135 
4136 /*
4137  * Function:
4138  *	vdc_handle_dring_reg_msg()
4139  *
4140  * Description:
4141  *
4142  * Arguments:
4143  *	vdc		- soft state pointer for this instance of the driver.
4144  *	dring_msg	- LDC message sent by vDisk server
4145  *
4146  * Return Code:
4147  *	0	- Success
4148  */
4149 static int
4150 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4151 {
4152 	int		status = 0;
4153 
4154 	ASSERT(vdc != NULL);
4155 	ASSERT(mutex_owned(&vdc->lock));
4156 
4157 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4158 		return (EPROTO);
4159 	}
4160 
4161 	switch (dring_msg->tag.vio_subtype) {
4162 	case VIO_SUBTYPE_ACK:
4163 		/* save the received dring_ident */
4164 		vdc->dring_ident = dring_msg->dring_ident;
4165 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4166 		    vdc->instance, vdc->dring_ident);
4167 		break;
4168 
4169 	case VIO_SUBTYPE_NACK:
4170 		/*
4171 		 * vds could not handle the DRing info we sent so we
4172 		 * stop negotiating.
4173 		 */
4174 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4175 		    vdc->instance);
4176 		status = EPROTO;
4177 		break;
4178 
4179 	case VIO_SUBTYPE_INFO:
4180 		/*
4181 		 * Handle the case where vds starts handshake
4182 		 * (for now only vdc is the instigatior)
4183 		 */
4184 		status = ENOTSUP;
4185 		break;
4186 	default:
4187 		status = ENOTSUP;
4188 	}
4189 
4190 	return (status);
4191 }
4192 
4193 /*
4194  * Function:
4195  *	vdc_verify_seq_num()
4196  *
4197  * Description:
4198  *	This functions verifies that the sequence number sent back by the vDisk
4199  *	server with the latest message is what is expected (i.e. it is greater
4200  *	than the last seq num sent by the vDisk server and less than or equal
4201  *	to the last seq num generated by vdc).
4202  *
4203  *	It then checks the request ID to see if any requests need processing
4204  *	in the DRing.
4205  *
4206  * Arguments:
4207  *	vdc		- soft state pointer for this instance of the driver.
4208  *	dring_msg	- pointer to the LDC message sent by vds
4209  *
4210  * Return Code:
4211  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4212  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4213  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4214  *				  vdc cannot deal with them
4215  */
4216 static int
4217 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
4218 {
4219 	ASSERT(vdc != NULL);
4220 	ASSERT(dring_msg != NULL);
4221 	ASSERT(mutex_owned(&vdc->lock));
4222 
4223 	/*
4224 	 * Check to see if the messages were responded to in the correct
4225 	 * order by vds.
4226 	 */
4227 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
4228 	    (dring_msg->seq_num > vdc->seq_num)) {
4229 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
4230 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
4231 		    vdc->instance, dring_msg->seq_num,
4232 		    vdc->seq_num_reply, vdc->seq_num,
4233 		    vdc->req_id_proc, vdc->req_id);
4234 		return (VDC_SEQ_NUM_INVALID);
4235 	}
4236 	vdc->seq_num_reply = dring_msg->seq_num;
4237 
4238 	if (vdc->req_id_proc < vdc->req_id)
4239 		return (VDC_SEQ_NUM_TODO);
4240 	else
4241 		return (VDC_SEQ_NUM_SKIP);
4242 }
4243 
4244 
4245 /*
4246  * Function:
4247  *	vdc_is_supported_version()
4248  *
4249  * Description:
4250  *	This routine checks if the major/minor version numbers specified in
4251  *	'ver_msg' are supported. If not it finds the next version that is
4252  *	in the supported version list 'vdc_version[]' and sets the fields in
4253  *	'ver_msg' to those values
4254  *
4255  * Arguments:
4256  *	ver_msg	- LDC message sent by vDisk server
4257  *
4258  * Return Code:
4259  *	B_TRUE	- Success
4260  *	B_FALSE	- Version not supported
4261  */
4262 static boolean_t
4263 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
4264 {
4265 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
4266 
4267 	for (int i = 0; i < vdc_num_versions; i++) {
4268 		ASSERT(vdc_version[i].major > 0);
4269 		ASSERT((i == 0) ||
4270 		    (vdc_version[i].major < vdc_version[i-1].major));
4271 
4272 		/*
4273 		 * If the major versions match, adjust the minor version, if
4274 		 * necessary, down to the highest value supported by this
4275 		 * client. The server should support all minor versions lower
4276 		 * than the value it sent
4277 		 */
4278 		if (ver_msg->ver_major == vdc_version[i].major) {
4279 			if (ver_msg->ver_minor > vdc_version[i].minor) {
4280 				DMSGX(0,
4281 				    "Adjusting minor version from %u to %u",
4282 				    ver_msg->ver_minor, vdc_version[i].minor);
4283 				ver_msg->ver_minor = vdc_version[i].minor;
4284 			}
4285 			return (B_TRUE);
4286 		}
4287 
4288 		/*
4289 		 * If the message contains a higher major version number, set
4290 		 * the message's major/minor versions to the current values
4291 		 * and return false, so this message will get resent with
4292 		 * these values, and the server will potentially try again
4293 		 * with the same or a lower version
4294 		 */
4295 		if (ver_msg->ver_major > vdc_version[i].major) {
4296 			ver_msg->ver_major = vdc_version[i].major;
4297 			ver_msg->ver_minor = vdc_version[i].minor;
4298 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
4299 			    ver_msg->ver_major, ver_msg->ver_minor);
4300 
4301 			return (B_FALSE);
4302 		}
4303 
4304 		/*
4305 		 * Otherwise, the message's major version is less than the
4306 		 * current major version, so continue the loop to the next
4307 		 * (lower) supported version
4308 		 */
4309 	}
4310 
4311 	/*
4312 	 * No common version was found; "ground" the version pair in the
4313 	 * message to terminate negotiation
4314 	 */
4315 	ver_msg->ver_major = 0;
4316 	ver_msg->ver_minor = 0;
4317 
4318 	return (B_FALSE);
4319 }
4320 /* -------------------------------------------------------------------------- */
4321 
4322 /*
4323  * DKIO(7) support
4324  */
4325 
4326 typedef struct vdc_dk_arg {
4327 	struct dk_callback	dkc;
4328 	int			mode;
4329 	dev_t			dev;
4330 	vdc_t			*vdc;
4331 } vdc_dk_arg_t;
4332 
4333 /*
4334  * Function:
4335  * 	vdc_dkio_flush_cb()
4336  *
4337  * Description:
4338  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
4339  *	by kernel code.
4340  *
4341  * Arguments:
4342  *	arg	- a pointer to a vdc_dk_arg_t structure.
4343  */
4344 void
4345 vdc_dkio_flush_cb(void *arg)
4346 {
4347 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
4348 	struct dk_callback	*dkc = NULL;
4349 	vdc_t			*vdc = NULL;
4350 	int			rv;
4351 
4352 	if (dk_arg == NULL) {
4353 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
4354 		return;
4355 	}
4356 	dkc = &dk_arg->dkc;
4357 	vdc = dk_arg->vdc;
4358 	ASSERT(vdc != NULL);
4359 
4360 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
4361 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir);
4362 	if (rv != 0) {
4363 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
4364 		    vdc->instance, rv,
4365 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
4366 	}
4367 
4368 	/*
4369 	 * Trigger the call back to notify the caller the the ioctl call has
4370 	 * been completed.
4371 	 */
4372 	if ((dk_arg->mode & FKIOCTL) &&
4373 	    (dkc != NULL) &&
4374 	    (dkc->dkc_callback != NULL)) {
4375 		ASSERT(dkc->dkc_cookie != NULL);
4376 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
4377 	}
4378 
4379 	/* Indicate that one less DKIO write flush is outstanding */
4380 	mutex_enter(&vdc->lock);
4381 	vdc->dkio_flush_pending--;
4382 	ASSERT(vdc->dkio_flush_pending >= 0);
4383 	mutex_exit(&vdc->lock);
4384 
4385 	/* free the mem that was allocated when the callback was dispatched */
4386 	kmem_free(arg, sizeof (vdc_dk_arg_t));
4387 }
4388 
4389 /*
4390  * Function:
4391  * 	vdc_dkio_get_partition()
4392  *
4393  * Description:
4394  *	This function implements the DKIOCGAPART ioctl.
4395  *
4396  * Arguments:
4397  *	dev	- device
4398  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
4399  *	flag	- ioctl flags
4400  */
4401 static int
4402 vdc_dkio_get_partition(dev_t dev, caddr_t arg, int flag)
4403 {
4404 	struct dk_geom geom;
4405 	struct vtoc vtoc;
4406 	union {
4407 		struct dk_map map[NDKMAP];
4408 		struct dk_map32 map32[NDKMAP];
4409 	} data;
4410 	int i, rv, size;
4411 
4412 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL);
4413 	if (rv != 0)
4414 		return (rv);
4415 
4416 	rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc, FKIOCTL);
4417 	if (rv != 0)
4418 		return (rv);
4419 
4420 	if (vtoc.v_nparts != NDKMAP ||
4421 	    geom.dkg_nhead == 0 || geom.dkg_nsect == 0)
4422 		return (EINVAL);
4423 
4424 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4425 
4426 		for (i = 0; i < NDKMAP; i++) {
4427 			data.map32[i].dkl_cylno = vtoc.v_part[i].p_start /
4428 			    (geom.dkg_nhead * geom.dkg_nsect);
4429 			data.map32[i].dkl_nblk = vtoc.v_part[i].p_size;
4430 		}
4431 		size = NDKMAP * sizeof (struct dk_map32);
4432 
4433 	} else {
4434 
4435 		for (i = 0; i < NDKMAP; i++) {
4436 			data.map[i].dkl_cylno = vtoc.v_part[i].p_start /
4437 			    (geom.dkg_nhead * geom.dkg_nsect);
4438 			data.map[i].dkl_nblk = vtoc.v_part[i].p_size;
4439 		}
4440 		size = NDKMAP * sizeof (struct dk_map);
4441 
4442 	}
4443 
4444 	if (ddi_copyout(&data, arg, size, flag) != 0)
4445 		return (EFAULT);
4446 
4447 	return (0);
4448 }
4449 
4450 /*
4451  * Function:
4452  * 	vdc_dioctl_rwcmd()
4453  *
4454  * Description:
4455  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
4456  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
4457  *
4458  * Arguments:
4459  *	dev	- device
4460  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
4461  *	flag	- ioctl flags
4462  */
4463 static int
4464 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
4465 {
4466 	struct dadkio_rwcmd32 rwcmd32;
4467 	struct dadkio_rwcmd rwcmd;
4468 	struct iovec aiov;
4469 	struct uio auio;
4470 	int rw, status;
4471 	struct buf *buf;
4472 
4473 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4474 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
4475 		    sizeof (struct dadkio_rwcmd32), flag)) {
4476 			return (EFAULT);
4477 		}
4478 		rwcmd.cmd = rwcmd32.cmd;
4479 		rwcmd.flags = rwcmd32.flags;
4480 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
4481 		rwcmd.buflen = rwcmd32.buflen;
4482 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
4483 	} else {
4484 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
4485 		    sizeof (struct dadkio_rwcmd), flag)) {
4486 			return (EFAULT);
4487 		}
4488 	}
4489 
4490 	switch (rwcmd.cmd) {
4491 	case DADKIO_RWCMD_READ:
4492 		rw = B_READ;
4493 		break;
4494 	case DADKIO_RWCMD_WRITE:
4495 		rw = B_WRITE;
4496 		break;
4497 	default:
4498 		return (EINVAL);
4499 	}
4500 
4501 	bzero((caddr_t)&aiov, sizeof (struct iovec));
4502 	aiov.iov_base   = rwcmd.bufaddr;
4503 	aiov.iov_len    = rwcmd.buflen;
4504 
4505 	bzero((caddr_t)&auio, sizeof (struct uio));
4506 	auio.uio_iov    = &aiov;
4507 	auio.uio_iovcnt = 1;
4508 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
4509 	auio.uio_resid  = rwcmd.buflen;
4510 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
4511 
4512 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
4513 	bioinit(buf);
4514 	/*
4515 	 * We use the private field of buf to specify that this is an
4516 	 * I/O using an absolute offset.
4517 	 */
4518 	buf->b_private = (void *)VD_SLICE_NONE;
4519 
4520 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
4521 
4522 	biofini(buf);
4523 	kmem_free(buf, sizeof (buf_t));
4524 
4525 	return (status);
4526 }
4527 
4528 /*
4529  * This structure is used in the DKIO(7I) array below.
4530  */
4531 typedef struct vdc_dk_ioctl {
4532 	uint8_t		op;		/* VD_OP_XXX value */
4533 	int		cmd;		/* Solaris ioctl operation number */
4534 	size_t		nbytes;		/* size of structure to be copied */
4535 
4536 	/* function to convert between vDisk and Solaris structure formats */
4537 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
4538 	    int mode, int dir);
4539 } vdc_dk_ioctl_t;
4540 
4541 /*
4542  * Subset of DKIO(7I) operations currently supported
4543  */
4544 static vdc_dk_ioctl_t	dk_ioctl[] = {
4545 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
4546 		vdc_null_copy_func},
4547 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
4548 		vdc_get_wce_convert},
4549 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
4550 		vdc_set_wce_convert},
4551 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
4552 		vdc_get_vtoc_convert},
4553 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
4554 		vdc_set_vtoc_convert},
4555 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
4556 		vdc_get_geom_convert},
4557 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
4558 		vdc_get_geom_convert},
4559 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
4560 		vdc_get_geom_convert},
4561 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
4562 		vdc_set_geom_convert},
4563 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
4564 		vdc_get_efi_convert},
4565 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
4566 		vdc_set_efi_convert},
4567 
4568 	/* DIOCTL_RWCMD is converted to a read or a write */
4569 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
4570 
4571 	/*
4572 	 * These particular ioctls are not sent to the server - vdc fakes up
4573 	 * the necessary info.
4574 	 */
4575 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
4576 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
4577 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
4578 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
4579 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
4580 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
4581 };
4582 
4583 /*
4584  * Function:
4585  *	vd_process_ioctl()
4586  *
4587  * Description:
4588  *	This routine processes disk specific ioctl calls
4589  *
4590  * Arguments:
4591  *	dev	- the device number
4592  *	cmd	- the operation [dkio(7I)] to be processed
4593  *	arg	- pointer to user provided structure
4594  *		  (contains data to be set or reference parameter for get)
4595  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
4596  *
4597  * Return Code:
4598  *	0
4599  *	EFAULT
4600  *	ENXIO
4601  *	EIO
4602  *	ENOTSUP
4603  */
4604 static int
4605 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode)
4606 {
4607 	int		instance = VDCUNIT(dev);
4608 	vdc_t		*vdc = NULL;
4609 	int		rv = -1;
4610 	int		idx = 0;		/* index into dk_ioctl[] */
4611 	size_t		len = 0;		/* #bytes to send to vds */
4612 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
4613 	caddr_t		mem_p = NULL;
4614 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
4615 	struct vtoc	vtoc_saved;
4616 	vdc_dk_ioctl_t	*iop;
4617 
4618 	vdc = ddi_get_soft_state(vdc_state, instance);
4619 	if (vdc == NULL) {
4620 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
4621 		    instance);
4622 		return (ENXIO);
4623 	}
4624 
4625 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
4626 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
4627 
4628 	/*
4629 	 * Validate the ioctl operation to be performed.
4630 	 *
4631 	 * If we have looped through the array without finding a match then we
4632 	 * don't support this ioctl.
4633 	 */
4634 	for (idx = 0; idx < nioctls; idx++) {
4635 		if (cmd == dk_ioctl[idx].cmd)
4636 			break;
4637 	}
4638 
4639 	if (idx >= nioctls) {
4640 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
4641 		    vdc->instance, cmd);
4642 		return (ENOTSUP);
4643 	}
4644 
4645 	iop = &(dk_ioctl[idx]);
4646 
4647 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
4648 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
4649 		dk_efi_t	dk_efi;
4650 
4651 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
4652 		if (rv != 0)
4653 			return (EFAULT);
4654 
4655 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
4656 	} else {
4657 		len = iop->nbytes;
4658 	}
4659 
4660 	/*
4661 	 * Deal with the ioctls which the server does not provide. vdc can
4662 	 * fake these up and return immediately
4663 	 */
4664 	switch (cmd) {
4665 	case CDROMREADOFFSET:
4666 	case DKIOCREMOVABLE:
4667 	case USCSICMD:
4668 		return (ENOTTY);
4669 
4670 	case DIOCTL_RWCMD:
4671 		{
4672 			if (vdc->cinfo->dki_ctype != DKC_DIRECT)
4673 				return (ENOTTY);
4674 
4675 			return (vdc_dioctl_rwcmd(dev, arg, mode));
4676 		}
4677 
4678 	case DKIOCGAPART:
4679 		{
4680 			if (vdc->vdisk_label != VD_DISK_LABEL_VTOC)
4681 				return (ENOTSUP);
4682 
4683 			return (vdc_dkio_get_partition(dev, arg, mode));
4684 		}
4685 
4686 	case DKIOCINFO:
4687 		{
4688 			struct dk_cinfo	cinfo;
4689 			if (vdc->cinfo == NULL)
4690 				return (ENXIO);
4691 
4692 			bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
4693 			cinfo.dki_partition = VDCPART(dev);
4694 
4695 			rv = ddi_copyout(&cinfo, (void *)arg,
4696 			    sizeof (struct dk_cinfo), mode);
4697 			if (rv != 0)
4698 				return (EFAULT);
4699 
4700 			return (0);
4701 		}
4702 
4703 	case DKIOCGMEDIAINFO:
4704 		{
4705 			if (vdc->minfo == NULL)
4706 				return (ENXIO);
4707 
4708 			rv = ddi_copyout(vdc->minfo, (void *)arg,
4709 			    sizeof (struct dk_minfo), mode);
4710 			if (rv != 0)
4711 				return (EFAULT);
4712 
4713 			return (0);
4714 		}
4715 
4716 	case DKIOCFLUSHWRITECACHE:
4717 		{
4718 			struct dk_callback *dkc = (struct dk_callback *)arg;
4719 			vdc_dk_arg_t	*dkarg = NULL;
4720 
4721 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
4722 			    instance, mode);
4723 
4724 			/*
4725 			 * If the backing device is not a 'real' disk then the
4726 			 * W$ operation request to the vDisk server will fail
4727 			 * so we might as well save the cycles and return now.
4728 			 */
4729 			if (vdc->vdisk_type != VD_DISK_TYPE_DISK)
4730 				return (ENOTTY);
4731 
4732 			/*
4733 			 * If arg is NULL, then there is no callback function
4734 			 * registered and the call operates synchronously; we
4735 			 * break and continue with the rest of the function and
4736 			 * wait for vds to return (i.e. after the request to
4737 			 * vds returns successfully, all writes completed prior
4738 			 * to the ioctl will have been flushed from the disk
4739 			 * write cache to persistent media.
4740 			 *
4741 			 * If a callback function is registered, we dispatch
4742 			 * the request on a task queue and return immediately.
4743 			 * The callback will deal with informing the calling
4744 			 * thread that the flush request is completed.
4745 			 */
4746 			if (dkc == NULL)
4747 				break;
4748 
4749 			/*
4750 			 * the asynchronous callback is only supported if
4751 			 * invoked from within the kernel
4752 			 */
4753 			if ((mode & FKIOCTL) == 0)
4754 				return (ENOTSUP);
4755 
4756 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
4757 
4758 			dkarg->mode = mode;
4759 			dkarg->dev = dev;
4760 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
4761 
4762 			mutex_enter(&vdc->lock);
4763 			vdc->dkio_flush_pending++;
4764 			dkarg->vdc = vdc;
4765 			mutex_exit(&vdc->lock);
4766 
4767 			/* put the request on a task queue */
4768 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
4769 			    (void *)dkarg, DDI_SLEEP);
4770 			if (rv == NULL) {
4771 				/* clean up if dispatch fails */
4772 				mutex_enter(&vdc->lock);
4773 				vdc->dkio_flush_pending--;
4774 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
4775 			}
4776 
4777 			return (rv == NULL ? ENOMEM : 0);
4778 		}
4779 	}
4780 
4781 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
4782 	ASSERT(iop->op != 0);
4783 
4784 	/* LDC requires that the memory being mapped is 8-byte aligned */
4785 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
4786 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
4787 	    instance, len, alloc_len);
4788 
4789 	ASSERT(alloc_len >= 0); /* sanity check */
4790 	if (alloc_len > 0)
4791 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
4792 
4793 	if (cmd == DKIOCSVTOC) {
4794 		/*
4795 		 * Save a copy of the current VTOC so that we can roll back
4796 		 * if the setting of the new VTOC fails.
4797 		 */
4798 		bcopy(vdc->vtoc, &vtoc_saved, sizeof (struct vtoc));
4799 	}
4800 
4801 	/*
4802 	 * Call the conversion function for this ioctl which, if necessary,
4803 	 * converts from the Solaris format to the format ARC'ed
4804 	 * as part of the vDisk protocol (FWARC 2006/195)
4805 	 */
4806 	ASSERT(iop->convert != NULL);
4807 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
4808 	if (rv != 0) {
4809 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
4810 		    instance, rv, cmd);
4811 		if (mem_p != NULL)
4812 			kmem_free(mem_p, alloc_len);
4813 		return (rv);
4814 	}
4815 
4816 	/*
4817 	 * send request to vds to service the ioctl.
4818 	 */
4819 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
4820 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
4821 	    VIO_both_dir);
4822 
4823 	if (rv != 0) {
4824 		/*
4825 		 * This is not necessarily an error. The ioctl could
4826 		 * be returning a value such as ENOTTY to indicate
4827 		 * that the ioctl is not applicable.
4828 		 */
4829 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
4830 		    instance, rv, cmd);
4831 		if (mem_p != NULL)
4832 			kmem_free(mem_p, alloc_len);
4833 
4834 		if (cmd == DKIOCSVTOC) {
4835 			/* update of the VTOC has failed, roll back */
4836 			bcopy(&vtoc_saved, vdc->vtoc, sizeof (struct vtoc));
4837 		}
4838 
4839 		return (rv);
4840 	}
4841 
4842 	if (cmd == DKIOCSVTOC) {
4843 		/*
4844 		 * The VTOC has been changed. We need to update the device
4845 		 * nodes to handle the case where an EFI label has been
4846 		 * changed to a VTOC label. We also try and update the device
4847 		 * node properties. Failing to set the properties should
4848 		 * not cause an error to be return the caller though.
4849 		 */
4850 		vdc->vdisk_label = VD_DISK_LABEL_VTOC;
4851 		(void) vdc_create_device_nodes_vtoc(vdc);
4852 
4853 		if (vdc_create_device_nodes_props(vdc)) {
4854 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
4855 			    " properties", vdc->instance);
4856 		}
4857 
4858 	} else if (cmd == DKIOCSETEFI) {
4859 		/*
4860 		 * The EFI has been changed. We need to update the device
4861 		 * nodes to handle the case where a VTOC label has been
4862 		 * changed to an EFI label. We also try and update the device
4863 		 * node properties. Failing to set the properties should
4864 		 * not cause an error to be return the caller though.
4865 		 */
4866 		struct dk_gpt *efi;
4867 		size_t efi_len;
4868 
4869 		vdc->vdisk_label = VD_DISK_LABEL_EFI;
4870 		(void) vdc_create_device_nodes_efi(vdc);
4871 
4872 		rv = vdc_efi_alloc_and_read(dev, &efi, &efi_len);
4873 
4874 		if (rv == 0) {
4875 			vdc_store_efi(vdc, efi);
4876 			rv = vdc_create_device_nodes_props(vdc);
4877 			vd_efi_free(efi, efi_len);
4878 		}
4879 
4880 		if (rv) {
4881 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
4882 			    " properties", vdc->instance);
4883 		}
4884 	}
4885 
4886 	/*
4887 	 * Call the conversion function (if it exists) for this ioctl
4888 	 * which converts from the format ARC'ed as part of the vDisk
4889 	 * protocol (FWARC 2006/195) back to a format understood by
4890 	 * the rest of Solaris.
4891 	 */
4892 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
4893 	if (rv != 0) {
4894 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
4895 		    instance, rv, cmd);
4896 		if (mem_p != NULL)
4897 			kmem_free(mem_p, alloc_len);
4898 		return (rv);
4899 	}
4900 
4901 	if (mem_p != NULL)
4902 		kmem_free(mem_p, alloc_len);
4903 
4904 	return (rv);
4905 }
4906 
4907 /*
4908  * Function:
4909  *
4910  * Description:
4911  *	This is an empty conversion function used by ioctl calls which
4912  *	do not need to convert the data being passed in/out to userland
4913  */
4914 static int
4915 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
4916 {
4917 	_NOTE(ARGUNUSED(vdc))
4918 	_NOTE(ARGUNUSED(from))
4919 	_NOTE(ARGUNUSED(to))
4920 	_NOTE(ARGUNUSED(mode))
4921 	_NOTE(ARGUNUSED(dir))
4922 
4923 	return (0);
4924 }
4925 
4926 static int
4927 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
4928     int mode, int dir)
4929 {
4930 	_NOTE(ARGUNUSED(vdc))
4931 
4932 	if (dir == VD_COPYIN)
4933 		return (0);		/* nothing to do */
4934 
4935 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
4936 		return (EFAULT);
4937 
4938 	return (0);
4939 }
4940 
4941 static int
4942 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
4943     int mode, int dir)
4944 {
4945 	_NOTE(ARGUNUSED(vdc))
4946 
4947 	if (dir == VD_COPYOUT)
4948 		return (0);		/* nothing to do */
4949 
4950 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
4951 		return (EFAULT);
4952 
4953 	return (0);
4954 }
4955 
4956 /*
4957  * Function:
4958  *	vdc_get_vtoc_convert()
4959  *
4960  * Description:
4961  *	This routine performs the necessary convertions from the DKIOCGVTOC
4962  *	Solaris structure to the format defined in FWARC 2006/195.
4963  *
4964  *	In the struct vtoc definition, the timestamp field is marked as not
4965  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
4966  *	However SVM uses that field to check it can write into the VTOC,
4967  *	so we fake up the info of that field.
4968  *
4969  * Arguments:
4970  *	vdc	- the vDisk client
4971  *	from	- the buffer containing the data to be copied from
4972  *	to	- the buffer to be copied to
4973  *	mode	- flags passed to ioctl() call
4974  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
4975  *
4976  * Return Code:
4977  *	0	- Success
4978  *	ENXIO	- incorrect buffer passed in.
4979  *	EFAULT	- ddi_copyout routine encountered an error.
4980  */
4981 static int
4982 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
4983 {
4984 	int		i;
4985 	void		*tmp_mem = NULL;
4986 	void		*tmp_memp;
4987 	struct vtoc	vt;
4988 	struct vtoc32	vt32;
4989 	int		copy_len = 0;
4990 	int		rv = 0;
4991 
4992 	if (dir != VD_COPYOUT)
4993 		return (0);	/* nothing to do */
4994 
4995 	if ((from == NULL) || (to == NULL))
4996 		return (ENXIO);
4997 
4998 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
4999 		copy_len = sizeof (struct vtoc32);
5000 	else
5001 		copy_len = sizeof (struct vtoc);
5002 
5003 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5004 
5005 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
5006 
5007 	/* fake the VTOC timestamp field */
5008 	for (i = 0; i < V_NUMPAR; i++) {
5009 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
5010 	}
5011 
5012 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5013 		vtoctovtoc32(vt, vt32);
5014 		tmp_memp = &vt32;
5015 	} else {
5016 		tmp_memp = &vt;
5017 	}
5018 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
5019 	if (rv != 0)
5020 		rv = EFAULT;
5021 
5022 	kmem_free(tmp_mem, copy_len);
5023 	return (rv);
5024 }
5025 
5026 /*
5027  * Function:
5028  *	vdc_set_vtoc_convert()
5029  *
5030  * Description:
5031  *	This routine performs the necessary convertions from the DKIOCSVTOC
5032  *	Solaris structure to the format defined in FWARC 2006/195.
5033  *
5034  * Arguments:
5035  *	vdc	- the vDisk client
5036  *	from	- Buffer with data
5037  *	to	- Buffer where data is to be copied to
5038  *	mode	- flags passed to ioctl
5039  *	dir	- direction of copy (in or out)
5040  *
5041  * Return Code:
5042  *	0	- Success
5043  *	ENXIO	- Invalid buffer passed in
5044  *	EFAULT	- ddi_copyin of data failed
5045  */
5046 static int
5047 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5048 {
5049 	void		*tmp_mem = NULL;
5050 	struct vtoc	vt;
5051 	struct vtoc	*vtp = &vt;
5052 	vd_vtoc_t	vtvd;
5053 	int		copy_len = 0;
5054 	int		rv = 0;
5055 
5056 	if (dir != VD_COPYIN)
5057 		return (0);	/* nothing to do */
5058 
5059 	if ((from == NULL) || (to == NULL))
5060 		return (ENXIO);
5061 
5062 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
5063 		copy_len = sizeof (struct vtoc32);
5064 	else
5065 		copy_len = sizeof (struct vtoc);
5066 
5067 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5068 
5069 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
5070 	if (rv != 0) {
5071 		kmem_free(tmp_mem, copy_len);
5072 		return (EFAULT);
5073 	}
5074 
5075 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5076 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
5077 	} else {
5078 		vtp = tmp_mem;
5079 	}
5080 
5081 	/*
5082 	 * The VTOC is being changed, then vdc needs to update the copy
5083 	 * it saved in the soft state structure.
5084 	 */
5085 	bcopy(vtp, vdc->vtoc, sizeof (struct vtoc));
5086 
5087 	VTOC2VD_VTOC(vtp, &vtvd);
5088 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
5089 	kmem_free(tmp_mem, copy_len);
5090 
5091 	return (0);
5092 }
5093 
5094 /*
5095  * Function:
5096  *	vdc_get_geom_convert()
5097  *
5098  * Description:
5099  *	This routine performs the necessary convertions from the DKIOCGGEOM,
5100  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
5101  *	defined in FWARC 2006/195
5102  *
5103  * Arguments:
5104  *	vdc	- the vDisk client
5105  *	from	- Buffer with data
5106  *	to	- Buffer where data is to be copied to
5107  *	mode	- flags passed to ioctl
5108  *	dir	- direction of copy (in or out)
5109  *
5110  * Return Code:
5111  *	0	- Success
5112  *	ENXIO	- Invalid buffer passed in
5113  *	EFAULT	- ddi_copyout of data failed
5114  */
5115 static int
5116 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5117 {
5118 	_NOTE(ARGUNUSED(vdc))
5119 
5120 	struct dk_geom	geom;
5121 	int	copy_len = sizeof (struct dk_geom);
5122 	int	rv = 0;
5123 
5124 	if (dir != VD_COPYOUT)
5125 		return (0);	/* nothing to do */
5126 
5127 	if ((from == NULL) || (to == NULL))
5128 		return (ENXIO);
5129 
5130 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
5131 	rv = ddi_copyout(&geom, to, copy_len, mode);
5132 	if (rv != 0)
5133 		rv = EFAULT;
5134 
5135 	return (rv);
5136 }
5137 
5138 /*
5139  * Function:
5140  *	vdc_set_geom_convert()
5141  *
5142  * Description:
5143  *	This routine performs the necessary convertions from the DKIOCSGEOM
5144  *	Solaris structure to the format defined in FWARC 2006/195.
5145  *
5146  * Arguments:
5147  *	vdc	- the vDisk client
5148  *	from	- Buffer with data
5149  *	to	- Buffer where data is to be copied to
5150  *	mode	- flags passed to ioctl
5151  *	dir	- direction of copy (in or out)
5152  *
5153  * Return Code:
5154  *	0	- Success
5155  *	ENXIO	- Invalid buffer passed in
5156  *	EFAULT	- ddi_copyin of data failed
5157  */
5158 static int
5159 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5160 {
5161 	_NOTE(ARGUNUSED(vdc))
5162 
5163 	vd_geom_t	vdgeom;
5164 	void		*tmp_mem = NULL;
5165 	int		copy_len = sizeof (struct dk_geom);
5166 	int		rv = 0;
5167 
5168 	if (dir != VD_COPYIN)
5169 		return (0);	/* nothing to do */
5170 
5171 	if ((from == NULL) || (to == NULL))
5172 		return (ENXIO);
5173 
5174 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5175 
5176 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
5177 	if (rv != 0) {
5178 		kmem_free(tmp_mem, copy_len);
5179 		return (EFAULT);
5180 	}
5181 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
5182 	bcopy(&vdgeom, to, sizeof (vdgeom));
5183 	kmem_free(tmp_mem, copy_len);
5184 
5185 	return (0);
5186 }
5187 
5188 static int
5189 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5190 {
5191 	_NOTE(ARGUNUSED(vdc))
5192 
5193 	vd_efi_t	*vd_efi;
5194 	dk_efi_t	dk_efi;
5195 	int		rv = 0;
5196 	void		*uaddr;
5197 
5198 	if ((from == NULL) || (to == NULL))
5199 		return (ENXIO);
5200 
5201 	if (dir == VD_COPYIN) {
5202 
5203 		vd_efi = (vd_efi_t *)to;
5204 
5205 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
5206 		if (rv != 0)
5207 			return (EFAULT);
5208 
5209 		vd_efi->lba = dk_efi.dki_lba;
5210 		vd_efi->length = dk_efi.dki_length;
5211 		bzero(vd_efi->data, vd_efi->length);
5212 
5213 	} else {
5214 
5215 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
5216 		if (rv != 0)
5217 			return (EFAULT);
5218 
5219 		uaddr = dk_efi.dki_data;
5220 
5221 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
5222 
5223 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
5224 
5225 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
5226 		    mode);
5227 		if (rv != 0)
5228 			return (EFAULT);
5229 
5230 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
5231 	}
5232 
5233 	return (0);
5234 }
5235 
5236 static int
5237 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5238 {
5239 	_NOTE(ARGUNUSED(vdc))
5240 
5241 	dk_efi_t	dk_efi;
5242 	void		*uaddr;
5243 
5244 	if (dir == VD_COPYOUT)
5245 		return (0);	/* nothing to do */
5246 
5247 	if ((from == NULL) || (to == NULL))
5248 		return (ENXIO);
5249 
5250 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
5251 		return (EFAULT);
5252 
5253 	uaddr = dk_efi.dki_data;
5254 
5255 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
5256 
5257 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
5258 		return (EFAULT);
5259 
5260 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
5261 
5262 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
5263 
5264 	return (0);
5265 }
5266 
5267 /*
5268  * Function:
5269  *	vdc_create_fake_geometry()
5270  *
5271  * Description:
5272  *	This routine fakes up the disk info needed for some DKIO ioctls.
5273  *		- DKIOCINFO
5274  *		- DKIOCGMEDIAINFO
5275  *
5276  *	[ just like lofi(7D) and ramdisk(7D) ]
5277  *
5278  * Arguments:
5279  *	vdc	- soft state pointer for this instance of the device driver.
5280  *
5281  * Return Code:
5282  *	0	- Success
5283  */
5284 static int
5285 vdc_create_fake_geometry(vdc_t *vdc)
5286 {
5287 	ASSERT(vdc != NULL);
5288 
5289 	/*
5290 	 * Check if max_xfer_sz and vdisk_size are valid
5291 	 */
5292 	if (vdc->vdisk_size == 0 || vdc->max_xfer_sz == 0)
5293 		return (EIO);
5294 
5295 	/*
5296 	 * DKIOCINFO support
5297 	 */
5298 	vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
5299 
5300 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
5301 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
5302 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
5303 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
5304 	/*
5305 	 * We currently set the controller type to DKC_DIRECT for any disk.
5306 	 * When SCSI support is implemented, we will eventually change this
5307 	 * type to DKC_SCSI_CCS for disks supporting the SCSI protocol.
5308 	 */
5309 	vdc->cinfo->dki_ctype = DKC_DIRECT;
5310 	vdc->cinfo->dki_flags = DKI_FMTVOL;
5311 	vdc->cinfo->dki_cnum = 0;
5312 	vdc->cinfo->dki_addr = 0;
5313 	vdc->cinfo->dki_space = 0;
5314 	vdc->cinfo->dki_prio = 0;
5315 	vdc->cinfo->dki_vec = 0;
5316 	vdc->cinfo->dki_unit = vdc->instance;
5317 	vdc->cinfo->dki_slave = 0;
5318 	/*
5319 	 * The partition number will be created on the fly depending on the
5320 	 * actual slice (i.e. minor node) that is used to request the data.
5321 	 */
5322 	vdc->cinfo->dki_partition = 0;
5323 
5324 	/*
5325 	 * DKIOCGMEDIAINFO support
5326 	 */
5327 	if (vdc->minfo == NULL)
5328 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
5329 	vdc->minfo->dki_media_type = DK_FIXED_DISK;
5330 	vdc->minfo->dki_capacity = vdc->vdisk_size;
5331 	vdc->minfo->dki_lbsize = DEV_BSIZE;
5332 
5333 	return (0);
5334 }
5335 
5336 /*
5337  * Function:
5338  *	vdc_setup_disk_layout()
5339  *
5340  * Description:
5341  *	This routine discovers all the necessary details about the "disk"
5342  *	by requesting the data that is available from the vDisk server and by
5343  *	faking up the rest of the data.
5344  *
5345  * Arguments:
5346  *	vdc	- soft state pointer for this instance of the device driver.
5347  *
5348  * Return Code:
5349  *	0	- Success
5350  */
5351 static int
5352 vdc_setup_disk_layout(vdc_t *vdc)
5353 {
5354 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
5355 	dev_t	dev;
5356 	int	slice = 0;
5357 	int	rv, error;
5358 
5359 	ASSERT(vdc != NULL);
5360 
5361 	if (vdc->vtoc == NULL)
5362 		vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
5363 
5364 	dev = makedevice(ddi_driver_major(vdc->dip),
5365 	    VD_MAKE_DEV(vdc->instance, 0));
5366 	rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)vdc->vtoc, FKIOCTL);
5367 
5368 	if (rv && rv != ENOTSUP) {
5369 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
5370 		    vdc->instance, rv);
5371 		return (rv);
5372 	}
5373 
5374 	/*
5375 	 * The process of attempting to read VTOC will initiate
5376 	 * the handshake and establish a connection. Following
5377 	 * handshake, go ahead and create geometry.
5378 	 */
5379 	error = vdc_create_fake_geometry(vdc);
5380 	if (error != 0) {
5381 		DMSG(vdc, 0, "[%d] Failed to create disk geometry (err%d)",
5382 		    vdc->instance, error);
5383 		return (error);
5384 	}
5385 
5386 	if (rv == ENOTSUP) {
5387 		/*
5388 		 * If the device does not support VTOC then we try
5389 		 * to read an EFI label.
5390 		 */
5391 		struct dk_gpt *efi;
5392 		size_t efi_len;
5393 
5394 		rv = vdc_efi_alloc_and_read(dev, &efi, &efi_len);
5395 
5396 		if (rv) {
5397 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
5398 			    vdc->instance, rv);
5399 			return (rv);
5400 		}
5401 
5402 		vdc->vdisk_label = VD_DISK_LABEL_EFI;
5403 		vdc_store_efi(vdc, efi);
5404 		vd_efi_free(efi, efi_len);
5405 
5406 		return (0);
5407 	}
5408 
5409 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
5410 
5411 	/*
5412 	 * FUTURE: This could be default way for reading the VTOC
5413 	 * from the disk as supposed to sending the VD_OP_GET_VTOC
5414 	 * to the server. Currently this is a sanity check.
5415 	 *
5416 	 * find the slice that represents the entire "disk" and use that to
5417 	 * read the disk label. The convention in Solaris is that slice 2
5418 	 * represents the whole disk so we check that it is, otherwise we
5419 	 * default to slice 0
5420 	 */
5421 	if ((vdc->vdisk_type == VD_DISK_TYPE_DISK) &&
5422 	    (vdc->vtoc->v_part[2].p_tag == V_BACKUP)) {
5423 		slice = 2;
5424 	} else {
5425 		slice = 0;
5426 	}
5427 
5428 	/*
5429 	 * Read disk label from start of disk
5430 	 */
5431 	vdc->label = kmem_zalloc(DK_LABEL_SIZE, KM_SLEEP);
5432 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5433 	bioinit(buf);
5434 	buf->b_un.b_addr = (caddr_t)vdc->label;
5435 	buf->b_bcount = DK_LABEL_SIZE;
5436 	buf->b_flags = B_BUSY | B_READ;
5437 	buf->b_dev = dev;
5438 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)vdc->label,
5439 	    DK_LABEL_SIZE, slice, 0, CB_STRATEGY, buf, VIO_read_dir);
5440 	if (rv) {
5441 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
5442 		    vdc->instance);
5443 		kmem_free(buf, sizeof (buf_t));
5444 		return (rv);
5445 	}
5446 	rv = biowait(buf);
5447 	biofini(buf);
5448 	kmem_free(buf, sizeof (buf_t));
5449 
5450 	return (rv);
5451 }
5452 
5453 /*
5454  * Function:
5455  *	vdc_setup_devid()
5456  *
5457  * Description:
5458  *	This routine discovers the devid of a vDisk. It requests the devid of
5459  *	the underlying device from the vDisk server, builds an encapsulated
5460  *	devid based on the retrieved devid and registers that new devid to
5461  *	the vDisk.
5462  *
5463  * Arguments:
5464  *	vdc	- soft state pointer for this instance of the device driver.
5465  *
5466  * Return Code:
5467  *	0	- A devid was succesfully registered for the vDisk
5468  */
5469 static int
5470 vdc_setup_devid(vdc_t *vdc)
5471 {
5472 	int rv;
5473 	vd_devid_t *vd_devid;
5474 	size_t bufsize, bufid_len;
5475 
5476 	/*
5477 	 * At first sight, we don't know the size of the devid that the
5478 	 * server will return but this size will be encoded into the
5479 	 * reply. So we do a first request using a default size then we
5480 	 * check if this size was large enough. If not then we do a second
5481 	 * request with the correct size returned by the server. Note that
5482 	 * ldc requires size to be 8-byte aligned.
5483 	 */
5484 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
5485 	    sizeof (uint64_t));
5486 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
5487 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
5488 
5489 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
5490 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir);
5491 
5492 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
5493 
5494 	if (rv) {
5495 		kmem_free(vd_devid, bufsize);
5496 		return (rv);
5497 	}
5498 
5499 	if (vd_devid->length > bufid_len) {
5500 		/*
5501 		 * The returned devid is larger than the buffer used. Try again
5502 		 * with a buffer with the right size.
5503 		 */
5504 		kmem_free(vd_devid, bufsize);
5505 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
5506 		    sizeof (uint64_t));
5507 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
5508 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
5509 
5510 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
5511 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
5512 		    VIO_both_dir);
5513 
5514 		if (rv) {
5515 			kmem_free(vd_devid, bufsize);
5516 			return (rv);
5517 		}
5518 	}
5519 
5520 	/*
5521 	 * The virtual disk should have the same device id as the one associated
5522 	 * with the physical disk it is mapped on, otherwise sharing a disk
5523 	 * between a LDom and a non-LDom may not work (for example for a shared
5524 	 * SVM disk set).
5525 	 *
5526 	 * The DDI framework does not allow creating a device id with any
5527 	 * type so we first create a device id of type DEVID_ENCAP and then
5528 	 * we restore the orignal type of the physical device.
5529 	 */
5530 
5531 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
5532 
5533 	/* build an encapsulated devid based on the returned devid */
5534 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
5535 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
5536 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
5537 		kmem_free(vd_devid, bufsize);
5538 		return (1);
5539 	}
5540 
5541 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
5542 
5543 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
5544 
5545 	kmem_free(vd_devid, bufsize);
5546 
5547 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
5548 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
5549 		return (1);
5550 	}
5551 
5552 	return (0);
5553 }
5554 
5555 static void
5556 vdc_store_efi(vdc_t *vdc, struct dk_gpt *efi)
5557 {
5558 	struct vtoc *vtoc = vdc->vtoc;
5559 
5560 	vd_efi_to_vtoc(efi, vtoc);
5561 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5562 		/*
5563 		 * vd_efi_to_vtoc() will store information about the EFI Sun
5564 		 * reserved partition (representing the entire disk) into
5565 		 * partition 7. However single-slice device will only have
5566 		 * that single partition and the vdc driver expects to find
5567 		 * information about that partition in slice 0. So we need
5568 		 * to copy information from slice 7 to slice 0.
5569 		 */
5570 		vtoc->v_part[0].p_tag = vtoc->v_part[VD_EFI_WD_SLICE].p_tag;
5571 		vtoc->v_part[0].p_flag = vtoc->v_part[VD_EFI_WD_SLICE].p_flag;
5572 		vtoc->v_part[0].p_start = vtoc->v_part[VD_EFI_WD_SLICE].p_start;
5573 		vtoc->v_part[0].p_size =  vtoc->v_part[VD_EFI_WD_SLICE].p_size;
5574 	}
5575 }
5576